Merge remote-tracking branch 'upstream/dev' into 20260218-zigbee-proxy

tests/benchmarks/components/.gitignore

@@ -0,0 +1,2 @@
+/.esphome/
+/secrets.yaml

tests/benchmarks/components/api/__init__.py

@@ -0,0 +1,29 @@
+import esphome.codegen as cg
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # api must run its to_code to define USE_API, USE_API_PLAINTEXT,
+    # and add the noise-c library dependency.
+    manifest.enable_codegen()
+
+    original_to_code = manifest.to_code
+
+    async def to_code(config):
+        await original_to_code(config)
+        # Enable proxy proto message types for benchmarks.  The real
+        # components have hardware dependencies (BLE/UART/RMT); lightweight
+        # stub headers in tests/benchmarks/stubs/ satisfy the includes.
+        cg.add_define("USE_BLUETOOTH_PROXY")
+        cg.add_define("BLUETOOTH_PROXY_MAX_CONNECTIONS", 3)
+        cg.add_define("BLUETOOTH_PROXY_ADVERTISEMENT_BATCH_SIZE", 16)
+        cg.add_define("USE_ZWAVE_PROXY")
+        cg.add_define("USE_INFRARED")
+        cg.add_define("USE_IR_RF")
+        cg.add_define("USE_RADIO_FREQUENCY")
+        cg.add_define("USE_SERIAL_PROXY")
+        cg.add_define("SERIAL_PROXY_COUNT", 0)
+        cg.add_define("ESPHOME_ENTITY_INFRARED_COUNT", 0)
+        cg.add_define("ESPHOME_ENTITY_RADIO_FREQUENCY_COUNT", 0)
+
+    manifest.to_code = to_code

tests/benchmarks/components/api/bench_helpers.h

@@ -0,0 +1,67 @@
+#pragma once
+
+#include <fcntl.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+#include <memory>
+#include <utility>
+
+#include "esphome/components/socket/socket.h"
+
+namespace esphome::api::benchmarks {
+
+// Helper to drain accumulated data from the read side of a socket
+// to prevent the write side from blocking.
+inline void drain_socket(int fd) {
+  char buf[65536];
+  while (::read(fd, buf, sizeof(buf)) > 0) {
+  }
+}
+
+// Create a TCP loopback socket pair. Returns the write-side Socket
+// (wrapped for ESPHome) and the raw read-side fd for draining.
+// Both ends are non-blocking with 16MB buffers.
+inline std::pair<std::unique_ptr<socket::Socket>, int> create_tcp_loopback() {
+  // Create a TCP listener on loopback
+  int listen_fd = ::socket(AF_INET, SOCK_STREAM, 0);
+  int opt = 1;
+  ::setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt));
+
+  struct sockaddr_in addr {};
+  addr.sin_family = AF_INET;
+  addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+  addr.sin_port = 0;  // OS-assigned port
+  ::bind(listen_fd, reinterpret_cast<struct sockaddr *>(&addr), sizeof(addr));
+  ::listen(listen_fd, 1);
+
+  // Get the assigned port
+  socklen_t addr_len = sizeof(addr);
+  ::getsockname(listen_fd, reinterpret_cast<struct sockaddr *>(&addr), &addr_len);
+
+  // Connect from client side
+  int write_fd = ::socket(AF_INET, SOCK_STREAM, 0);
+  ::connect(write_fd, reinterpret_cast<struct sockaddr *>(&addr), sizeof(addr));
+
+  // Accept on server side (this is our read fd)
+  int read_fd = ::accept(listen_fd, nullptr, nullptr);
+  ::close(listen_fd);
+
+  // Make both ends non-blocking
+  int flags = ::fcntl(write_fd, F_GETFL, 0);
+  ::fcntl(write_fd, F_SETFL, flags | O_NONBLOCK);
+  flags = ::fcntl(read_fd, F_GETFL, 0);
+  ::fcntl(read_fd, F_SETFL, flags | O_NONBLOCK);
+
+  // Use large socket buffers so benchmarks never hit WOULD_BLOCK
+  // during a single outer iteration (2000 × ~15B messages = ~30KB).
+  int bufsize = 16 * 1024 * 1024;
+  ::setsockopt(write_fd, SOL_SOCKET, SO_SNDBUF, &bufsize, sizeof(bufsize));
+  ::setsockopt(read_fd, SOL_SOCKET, SO_RCVBUF, &bufsize, sizeof(bufsize));
+
+  return {std::make_unique<socket::Socket>(write_fd), read_fd};
+}
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_list_entities.cpp

@@ -0,0 +1,235 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+#include "esphome/components/light/color_mode.h"
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// --- ListEntitiesSensorResponse ---
+
+static ListEntitiesSensorResponse make_sensor_response() {
+  ListEntitiesSensorResponse msg;
+  msg.object_id = StringRef::from_lit("living_room_temperature");
+  msg.key = 0x12345678;
+  msg.name = StringRef::from_lit("Living Room Temperature");
+#ifdef USE_ENTITY_ICON
+  msg.icon = StringRef::from_lit("mdi:thermometer");
+#endif
+  msg.entity_category = enums::ENTITY_CATEGORY_NONE;
+  msg.disabled_by_default = false;
+  msg.unit_of_measurement = StringRef::from_lit("°C");
+  msg.accuracy_decimals = 1;
+  msg.force_update = false;
+  msg.device_class = StringRef::from_lit("temperature");
+  msg.state_class = enums::STATE_CLASS_MEASUREMENT;
+#ifdef USE_DEVICES
+  msg.device_id = 1;
+#endif
+  return msg;
+}
+
+static void CalculateSize_ListEntitiesSensorResponse(benchmark::State &state) {
+  auto msg = make_sensor_response();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_ListEntitiesSensorResponse);
+
+static void Encode_ListEntitiesSensorResponse(benchmark::State &state) {
+  auto msg = make_sensor_response();
+  APIBuffer buffer;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_ListEntitiesSensorResponse);
+
+static void CalcAndEncode_ListEntitiesSensorResponse(benchmark::State &state) {
+  auto msg = make_sensor_response();
+  APIBuffer buffer;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_ListEntitiesSensorResponse);
+
+// --- ListEntitiesBinarySensorResponse ---
+
+static ListEntitiesBinarySensorResponse make_binary_sensor_response() {
+  ListEntitiesBinarySensorResponse msg;
+  msg.object_id = StringRef::from_lit("front_door_contact");
+  msg.key = 0xAABBCCDD;
+  msg.name = StringRef::from_lit("Front Door Contact");
+#ifdef USE_ENTITY_ICON
+  msg.icon = StringRef::from_lit("mdi:door");
+#endif
+  msg.entity_category = enums::ENTITY_CATEGORY_NONE;
+  msg.disabled_by_default = false;
+  msg.device_class = StringRef::from_lit("door");
+  msg.is_status_binary_sensor = false;
+#ifdef USE_DEVICES
+  msg.device_id = 2;
+#endif
+  return msg;
+}
+
+static void CalculateSize_ListEntitiesBinarySensorResponse(benchmark::State &state) {
+  auto msg = make_binary_sensor_response();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_ListEntitiesBinarySensorResponse);
+
+static void Encode_ListEntitiesBinarySensorResponse(benchmark::State &state) {
+  auto msg = make_binary_sensor_response();
+  APIBuffer buffer;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_ListEntitiesBinarySensorResponse);
+
+static void CalcAndEncode_ListEntitiesBinarySensorResponse(benchmark::State &state) {
+  auto msg = make_binary_sensor_response();
+  APIBuffer buffer;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_ListEntitiesBinarySensorResponse);
+
+// --- ListEntitiesLightResponse ---
+
+static light::ColorModeMask light_color_modes;
+static FixedVector<const char *> light_effects;
+
+static ListEntitiesLightResponse make_light_response() {
+  // Initialize static data on first call
+  static bool initialized = false;
+  if (!initialized) {
+    light_color_modes.insert(light::ColorMode::RGB_WHITE);
+    light_color_modes.insert(light::ColorMode::COLOR_TEMPERATURE);
+    light_effects.init(3);
+    light_effects.push_back("None");
+    light_effects.push_back("Rainbow");
+    light_effects.push_back("Strobe");
+    initialized = true;
+  }
+
+  ListEntitiesLightResponse msg;
+  msg.object_id = StringRef::from_lit("kitchen_ceiling_light");
+  msg.key = 0x55667788;
+  msg.name = StringRef::from_lit("Kitchen Ceiling Light");
+#ifdef USE_ENTITY_ICON
+  msg.icon = StringRef::from_lit("mdi:ceiling-light");
+#endif
+  msg.entity_category = enums::ENTITY_CATEGORY_NONE;
+  msg.disabled_by_default = false;
+  msg.supported_color_modes = &light_color_modes;
+  msg.min_mireds = 153.0f;
+  msg.max_mireds = 500.0f;
+  msg.effects = &light_effects;
+#ifdef USE_DEVICES
+  msg.device_id = 3;
+#endif
+  return msg;
+}
+
+static void CalculateSize_ListEntitiesLightResponse(benchmark::State &state) {
+  auto msg = make_light_response();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_ListEntitiesLightResponse);
+
+static void Encode_ListEntitiesLightResponse(benchmark::State &state) {
+  auto msg = make_light_response();
+  APIBuffer buffer;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_ListEntitiesLightResponse);
+
+static void CalcAndEncode_ListEntitiesLightResponse(benchmark::State &state) {
+  auto msg = make_light_response();
+  APIBuffer buffer;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_ListEntitiesLightResponse);
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_log_response.cpp

@@ -0,0 +1,118 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Typical log line: "[12:34:56][D][sensor:094]: 'Temperature': Sending state 23.50000 with 1 decimals of accuracy"
+static constexpr const char *kTypicalLogLine =
+    "[12:34:56][D][sensor:094]: 'Temperature': Sending state 23.50000 with 1 decimals of accuracy";
+
+// Short log line: "[12:34:56][I][app:029]: Running..."
+static constexpr const char *kShortLogLine = "[12:34:56][I][app:029]: Running...";
+
+// --- Encode ---
+
+static void Encode_LogResponse_Typical(benchmark::State &state) {
+  APIBuffer buffer;
+  SubscribeLogsResponse msg;
+  msg.level = enums::LOG_LEVEL_DEBUG;
+  msg.set_message(reinterpret_cast<const uint8_t *>(kTypicalLogLine), strlen(kTypicalLogLine));
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_LogResponse_Typical);
+
+static void Encode_LogResponse_Short(benchmark::State &state) {
+  APIBuffer buffer;
+  SubscribeLogsResponse msg;
+  msg.level = enums::LOG_LEVEL_INFO;
+  msg.set_message(reinterpret_cast<const uint8_t *>(kShortLogLine), strlen(kShortLogLine));
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_LogResponse_Short);
+
+// --- Calculate Size ---
+
+static void CalculateSize_LogResponse_Typical(benchmark::State &state) {
+  SubscribeLogsResponse msg;
+  msg.level = enums::LOG_LEVEL_DEBUG;
+  msg.set_message(reinterpret_cast<const uint8_t *>(kTypicalLogLine), strlen(kTypicalLogLine));
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_LogResponse_Typical);
+
+// --- Calc + Encode (steady state) ---
+
+static void CalcAndEncode_LogResponse_Typical(benchmark::State &state) {
+  APIBuffer buffer;
+  SubscribeLogsResponse msg;
+  msg.level = enums::LOG_LEVEL_DEBUG;
+  msg.set_message(reinterpret_cast<const uint8_t *>(kTypicalLogLine), strlen(kTypicalLogLine));
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_LogResponse_Typical);
+
+// --- Calc + Encode (fresh allocation each time) ---
+
+static void CalcAndEncode_LogResponse_Typical_Fresh(benchmark::State &state) {
+  SubscribeLogsResponse msg;
+  msg.level = enums::LOG_LEVEL_DEBUG;
+  msg.set_message(reinterpret_cast<const uint8_t *>(kTypicalLogLine), strlen(kTypicalLogLine));
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      APIBuffer buffer;
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+      benchmark::DoNotOptimize(buffer.data());
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_LogResponse_Typical_Fresh);
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_noise_encrypt.cpp

@@ -0,0 +1,306 @@
+#include "esphome/core/defines.h"
+#ifdef USE_API_NOISE
+
+#include <benchmark/benchmark.h>
+#include <cstring>
+#include <memory>
+
+#include "noise/protocol.h"
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Helper to create and initialize a NoiseCipherState with ChaChaPoly.
+// Returns nullptr on failure.
+static NoiseCipherState *create_cipher() {
+  NoiseCipherState *cipher = nullptr;
+  int err = noise_cipherstate_new_by_id(&cipher, NOISE_CIPHER_CHACHAPOLY);
+  if (err != NOISE_ERROR_NONE || cipher == nullptr)
+    return nullptr;
+
+  // Initialize with a dummy 32-byte key (same pattern as handshake split produces)
+  uint8_t key[32];
+  memset(key, 0xAB, sizeof(key));
+  err = noise_cipherstate_init_key(cipher, key, sizeof(key));
+  if (err != NOISE_ERROR_NONE) {
+    noise_cipherstate_free(cipher);
+    return nullptr;
+  }
+  return cipher;
+}
+
+// Benchmark helper matching the exact pattern from
+// APINoiseFrameHelper::write_protobuf_messages:
+//   - noise_buffer_init + noise_buffer_set_inout (same as production)
+//   - No explicit set_nonce (production relies on internal nonce increment)
+//   - Error checking on encrypt return
+static void noise_encrypt_bench(benchmark::State &state, size_t plaintext_size) {
+  NoiseCipherState *cipher = create_cipher();
+  if (cipher == nullptr) {
+    state.SkipWithError("Failed to create cipher state");
+    return;
+  }
+
+  size_t mac_len = noise_cipherstate_get_mac_length(cipher);
+  size_t buf_capacity = plaintext_size + mac_len;
+  auto buffer = std::make_unique<uint8_t[]>(buf_capacity);
+  memset(buffer.get(), 0x42, plaintext_size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      // Match production: init buffer, set inout, encrypt
+      NoiseBuffer mbuf;
+      noise_buffer_init(mbuf);
+      noise_buffer_set_inout(mbuf, buffer.get(), plaintext_size, buf_capacity);
+
+      int err = noise_cipherstate_encrypt(cipher, &mbuf);
+      if (err != NOISE_ERROR_NONE) {
+        state.SkipWithError("noise_cipherstate_encrypt failed");
+        noise_cipherstate_free(cipher);
+        return;
+      }
+    }
+    benchmark::DoNotOptimize(buffer[0]);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  noise_cipherstate_free(cipher);
+}
+
+// --- Encrypt a typical sensor state message (small payload ~14 bytes) ---
+// This is the most common message encrypted on every sensor update.
+// 4 bytes type+len header + ~10 bytes payload.
+
+static void NoiseEncrypt_SmallMessage(benchmark::State &state) { noise_encrypt_bench(state, 14); }
+BENCHMARK(NoiseEncrypt_SmallMessage);
+
+// --- Encrypt a medium message (~128 bytes, typical for LightStateResponse) ---
+
+static void NoiseEncrypt_MediumMessage(benchmark::State &state) { noise_encrypt_bench(state, 128); }
+BENCHMARK(NoiseEncrypt_MediumMessage);
+
+// --- Encrypt a large message (~1024 bytes, typical for DeviceInfoResponse) ---
+
+static void NoiseEncrypt_LargeMessage(benchmark::State &state) { noise_encrypt_bench(state, 1024); }
+BENCHMARK(NoiseEncrypt_LargeMessage);
+
+// Benchmark helper matching the exact pattern from
+// APINoiseFrameHelper::read_packet:
+//   - noise_buffer_init + noise_buffer_set_inout with capacity == size (decrypt shrinks)
+//   - Error checking on decrypt return
+//
+// Pre-encrypts kInnerIterations messages with sequential nonces before the
+// timed loop. Each outer iteration re-inits the decrypt key to reset the
+// nonce back to 0, then decrypts all pre-encrypted messages in sequence.
+// The init_key cost is amortized over kInnerIterations decrypts.
+static void noise_decrypt_bench(benchmark::State &state, size_t plaintext_size) {
+  NoiseCipherState *encrypt_cipher = create_cipher();
+  NoiseCipherState *decrypt_cipher = create_cipher();
+  if (encrypt_cipher == nullptr || decrypt_cipher == nullptr) {
+    state.SkipWithError("Failed to create cipher state");
+    if (encrypt_cipher)
+      noise_cipherstate_free(encrypt_cipher);
+    if (decrypt_cipher)
+      noise_cipherstate_free(decrypt_cipher);
+    return;
+  }
+
+  size_t mac_len = noise_cipherstate_get_mac_length(encrypt_cipher);
+  size_t encrypted_size = plaintext_size + mac_len;
+
+  // Pre-encrypt kInnerIterations messages with sequential nonces (0..N-1).
+  auto ciphertexts = std::make_unique<uint8_t[]>(encrypted_size * kInnerIterations);
+  for (int i = 0; i < kInnerIterations; i++) {
+    uint8_t *ct = ciphertexts.get() + i * encrypted_size;
+    memset(ct, 0x42, plaintext_size);
+    NoiseBuffer enc_buf;
+    noise_buffer_init(enc_buf);
+    noise_buffer_set_inout(enc_buf, ct, plaintext_size, encrypted_size);
+    int err = noise_cipherstate_encrypt(encrypt_cipher, &enc_buf);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Pre-encrypt failed");
+      noise_cipherstate_free(encrypt_cipher);
+      noise_cipherstate_free(decrypt_cipher);
+      return;
+    }
+  }
+
+  // Working buffer — decrypt modifies in place
+  auto buffer = std::make_unique<uint8_t[]>(encrypted_size);
+  static constexpr uint8_t KEY[32] = {0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB,
+                                      0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB,
+                                      0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB};
+
+  for (auto _ : state) {
+    // Reset nonce to 0 by re-initing the key (amortized over kInnerIterations)
+    noise_cipherstate_init_key(decrypt_cipher, KEY, sizeof(KEY));
+
+    for (int i = 0; i < kInnerIterations; i++) {
+      // Copy ciphertext into working buffer (decrypt modifies in place)
+      memcpy(buffer.get(), ciphertexts.get() + i * encrypted_size, encrypted_size);
+
+      // Decrypt matching production pattern
+      NoiseBuffer mbuf;
+      noise_buffer_init(mbuf);
+      noise_buffer_set_inout(mbuf, buffer.get(), encrypted_size, encrypted_size);
+
+      int err = noise_cipherstate_decrypt(decrypt_cipher, &mbuf);
+      if (err != NOISE_ERROR_NONE) {
+        state.SkipWithError("noise_cipherstate_decrypt failed");
+        noise_cipherstate_free(encrypt_cipher);
+        noise_cipherstate_free(decrypt_cipher);
+        return;
+      }
+    }
+    benchmark::DoNotOptimize(buffer[0]);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  noise_cipherstate_free(encrypt_cipher);
+  noise_cipherstate_free(decrypt_cipher);
+}
+
+// --- Decrypt benchmarks (matching read_packet path) ---
+
+static void NoiseDecrypt_SmallMessage(benchmark::State &state) { noise_decrypt_bench(state, 14); }
+BENCHMARK(NoiseDecrypt_SmallMessage);
+
+static void NoiseDecrypt_MediumMessage(benchmark::State &state) { noise_decrypt_bench(state, 128); }
+BENCHMARK(NoiseDecrypt_MediumMessage);
+
+static void NoiseDecrypt_LargeMessage(benchmark::State &state) { noise_decrypt_bench(state, 1024); }
+BENCHMARK(NoiseDecrypt_LargeMessage);
+
+// --- Full Noise_NNpsk0 handshake benchmark ---
+// Measures the complete handshake between initiator and responder:
+//   - Create handshake states for both sides
+//   - Set PSK and prologue
+//   - Exchange messages (initiator write -> responder read -> responder write -> initiator read)
+//   - Split to get cipher states
+// This is dominated by Curve25519 DH operations (expensive on ESP8266).
+// No inner iterations — each handshake is already expensive enough.
+
+static void NoiseHandshake_Full(benchmark::State &state) {
+  // Matching ESPHome's protocol: Noise_NNpsk0_25519_ChaChaPoly_SHA256
+  NoiseProtocolId nid;
+  memset(&nid, 0, sizeof(nid));
+  nid.pattern_id = NOISE_PATTERN_NN;
+  nid.cipher_id = NOISE_CIPHER_CHACHAPOLY;
+  nid.dh_id = NOISE_DH_CURVE25519;
+  nid.prefix_id = NOISE_PREFIX_STANDARD;
+  nid.hybrid_id = NOISE_DH_NONE;
+  nid.hash_id = NOISE_HASH_SHA256;
+  nid.modifier_ids[0] = NOISE_MODIFIER_PSK0;
+
+  // Dummy PSK (32 bytes) and prologue matching production setup
+  static constexpr uint8_t PSK[32] = {0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB,
+                                      0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB,
+                                      0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB, 0xAB};
+  static constexpr uint8_t PROLOGUE[] = "NoESPHome";
+
+  // Message buffer for handshake exchange (max handshake message ~96 bytes)
+  uint8_t msg_buf[128];
+
+  for (auto _ : state) {
+    NoiseHandshakeState *initiator = nullptr;
+    NoiseHandshakeState *responder = nullptr;
+    NoiseCipherState *init_send = nullptr, *init_recv = nullptr;
+    NoiseCipherState *resp_send = nullptr, *resp_recv = nullptr;
+    int err;
+
+    // Create both handshake states
+    err = noise_handshakestate_new_by_id(&initiator, &nid, NOISE_ROLE_INITIATOR);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Failed to create initiator");
+      return;
+    }
+    err = noise_handshakestate_new_by_id(&responder, &nid, NOISE_ROLE_RESPONDER);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Failed to create responder");
+      noise_handshakestate_free(initiator);
+      return;
+    }
+
+    // Set PSK and prologue on both sides
+    noise_handshakestate_set_pre_shared_key(initiator, PSK, sizeof(PSK));
+    noise_handshakestate_set_pre_shared_key(responder, PSK, sizeof(PSK));
+    noise_handshakestate_set_prologue(initiator, PROLOGUE, sizeof(PROLOGUE) - 1);
+    noise_handshakestate_set_prologue(responder, PROLOGUE, sizeof(PROLOGUE) - 1);
+
+    noise_handshakestate_start(initiator);
+    noise_handshakestate_start(responder);
+
+    // Message 1: Initiator -> Responder
+    NoiseBuffer write_buf, read_buf;
+    noise_buffer_set_output(write_buf, msg_buf, sizeof(msg_buf));
+    err = noise_handshakestate_write_message(initiator, &write_buf, nullptr);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Initiator write_message failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      return;
+    }
+
+    noise_buffer_set_input(read_buf, msg_buf, write_buf.size);
+    err = noise_handshakestate_read_message(responder, &read_buf, nullptr);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Responder read_message failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      return;
+    }
+
+    // Message 2: Responder -> Initiator
+    noise_buffer_set_output(write_buf, msg_buf, sizeof(msg_buf));
+    err = noise_handshakestate_write_message(responder, &write_buf, nullptr);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Responder write_message failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      return;
+    }
+
+    noise_buffer_set_input(read_buf, msg_buf, write_buf.size);
+    err = noise_handshakestate_read_message(initiator, &read_buf, nullptr);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Initiator read_message failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      return;
+    }
+
+    // Split to get cipher states
+    err = noise_handshakestate_split(initiator, &init_send, &init_recv);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Initiator split failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      return;
+    }
+    err = noise_handshakestate_split(responder, &resp_send, &resp_recv);
+    if (err != NOISE_ERROR_NONE) {
+      state.SkipWithError("Responder split failed");
+      noise_handshakestate_free(initiator);
+      noise_handshakestate_free(responder);
+      noise_cipherstate_free(init_send);
+      noise_cipherstate_free(init_recv);
+      return;
+    }
+
+    benchmark::DoNotOptimize(init_send);
+
+    // Cleanup
+    noise_handshakestate_free(initiator);
+    noise_handshakestate_free(responder);
+    noise_cipherstate_free(init_send);
+    noise_cipherstate_free(init_recv);
+    noise_cipherstate_free(resp_send);
+    noise_cipherstate_free(resp_recv);
+  }
+}
+BENCHMARK(NoiseHandshake_Full);
+
+}  // namespace esphome::api::benchmarks
+
+#endif  // USE_API_NOISE

tests/benchmarks/components/api/bench_plaintext_frame.cpp

@@ -0,0 +1,108 @@
+#include "esphome/core/defines.h"
+#ifdef USE_API_PLAINTEXT
+
+#include <benchmark/benchmark.h>
+#include <unistd.h>
+
+#include "bench_helpers.h"
+#include "esphome/components/api/api_frame_helper_plaintext.h"
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Helper to create a TCP loopback connection with an APIPlaintextFrameHelper
+// on the write end. Returns the helper and the read-side fd.
+static std::pair<std::unique_ptr<APIPlaintextFrameHelper>, int> create_plaintext_helper() {
+  auto [sock, read_fd] = create_tcp_loopback();
+  auto helper = std::make_unique<APIPlaintextFrameHelper>(std::move(sock));
+  helper->init();
+  return {std::move(helper), read_fd};
+}
+
+// --- Write a single SensorStateResponse through plaintext framing ---
+// Measures the full write path: header construction, varint encoding,
+// iovec assembly, and socket write.
+
+static void PlaintextFrame_WriteSensorState(benchmark::State &state) {
+  auto [helper, read_fd] = create_plaintext_helper();
+  uint8_t padding = helper->frame_header_padding();
+
+  // Pre-init buffer to typical TCP MSS size to avoid benchmarking
+  // heap allocation — in real use the buffer is reused across writes.
+  APIBuffer buffer;
+  buffer.reserve(1460);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      buffer.clear();
+      SensorStateResponse msg;
+      msg.key = 0x12345678;
+      msg.state = 23.5f;
+      msg.missing_state = false;
+
+      uint32_t size = msg.calculate_size();
+      buffer.resize(padding + size);
+      ProtoWriteBuffer writer(&buffer, padding);
+      msg.encode(writer);
+
+      helper->write_protobuf_packet(SensorStateResponse::MESSAGE_TYPE, writer);
+    }
+    drain_socket(read_fd);
+    benchmark::DoNotOptimize(helper.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(PlaintextFrame_WriteSensorState);
+
+// --- Write a batch of 5 SensorStateResponses in one call ---
+// Measures batched write: multiple messages assembled into one writev.
+
+static void PlaintextFrame_WriteBatch5(benchmark::State &state) {
+  auto [helper, read_fd] = create_plaintext_helper();
+  uint8_t padding = helper->frame_header_padding();
+  uint8_t footer = helper->frame_footer_size();
+
+  // Pre-init buffer to typical TCP MSS size to avoid benchmarking
+  // heap allocation — in real use the buffer is reused across writes.
+  APIBuffer buffer;
+  buffer.reserve(1460);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      buffer.clear();
+      MessageInfo messages[5] = {{0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}, {0, 0, 0, 0}};
+
+      for (int j = 0; j < 5; j++) {
+        uint16_t offset = buffer.size();
+        SensorStateResponse msg;
+        msg.key = static_cast<uint32_t>(j);
+        msg.state = 23.5f + static_cast<float>(j);
+        msg.missing_state = false;
+
+        uint32_t size = msg.calculate_size();
+        buffer.resize(offset + padding + size + footer);
+        ProtoWriteBuffer writer(&buffer, offset + padding);
+        msg.encode(writer);
+
+        messages[j] = MessageInfo(SensorStateResponse::MESSAGE_TYPE, offset, size, padding);
+      }
+
+      helper->write_protobuf_messages(ProtoWriteBuffer(&buffer, 0), std::span<const MessageInfo>(messages, 5));
+    }
+    drain_socket(read_fd);
+    benchmark::DoNotOptimize(helper.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(PlaintextFrame_WriteBatch5);
+
+}  // namespace esphome::api::benchmarks
+
+#endif  // USE_API_PLAINTEXT

tests/benchmarks/components/api/bench_proto_decode.cpp

@@ -0,0 +1,113 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+// Helper: encode a message into an APIBuffer for reuse in decode benchmarks.
+// Optimization barriers are applied to the decode target objects via
+// DoNotOptimize/ClobberMemory, not to this buffer.
+template<typename T> static APIBuffer encode_message(const T &msg) {
+  APIBuffer buffer;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+  ProtoWriteBuffer writer(&buffer, 0);
+  msg.encode(writer);
+  return buffer;
+}
+
+/// Force a pointer through an asm barrier so the compiler cannot
+/// prove its contents are unchanged across iterations.
+/// benchmark::DoNotOptimize/ClobberMemory are insufficient under
+/// CodSpeed's valgrind-based instrumentation.
+static void escape(void *p) { asm volatile("" : : "g"(p) : "memory"); }
+
+// --- HelloRequest decode (string + varint fields) ---
+
+static void Decode_HelloRequest(benchmark::State &state) {
+  HelloRequest source;
+  source.client_info = StringRef::from_lit("aioesphomeapi");
+  source.api_version_major = 1;
+  source.api_version_minor = 10;
+  auto encoded = encode_message(source);
+  auto *data = encoded.data();
+  auto size = encoded.size();
+  benchmark::DoNotOptimize(data);
+  benchmark::DoNotOptimize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      HelloRequest msg;
+      escape(&msg);
+      msg.decode(data, size);
+      escape(&msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_HelloRequest);
+
+// --- SwitchCommandRequest decode (simple command) ---
+
+static void Decode_SwitchCommandRequest(benchmark::State &state) {
+  SwitchCommandRequest source;
+  source.key = 0x12345678;
+  source.state = true;
+  auto encoded = encode_message(source);
+  auto *data = encoded.data();
+  auto size = encoded.size();
+  benchmark::DoNotOptimize(data);
+  benchmark::DoNotOptimize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      SwitchCommandRequest msg;
+      escape(&msg);
+      msg.decode(data, size);
+      escape(&msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_SwitchCommandRequest);
+
+// --- LightCommandRequest decode (complex command with many fields) ---
+
+static void Decode_LightCommandRequest(benchmark::State &state) {
+  LightCommandRequest source;
+  source.key = 0x11223344;
+  source.has_state = true;
+  source.state = true;
+  source.has_brightness = true;
+  source.brightness = 0.8f;
+  source.has_rgb = true;
+  source.red = 1.0f;
+  source.green = 0.5f;
+  source.blue = 0.2f;
+  source.has_effect = true;
+  source.effect = StringRef::from_lit("rainbow");
+  auto encoded = encode_message(source);
+  auto *data = encoded.data();
+  auto size = encoded.size();
+  benchmark::DoNotOptimize(data);
+  benchmark::DoNotOptimize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      LightCommandRequest msg;
+      escape(&msg);
+      msg.decode(data, size);
+      escape(&msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_LightCommandRequest);
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_proto_encode.cpp

@@ -0,0 +1,387 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+// --- SensorStateResponse (highest frequency message) ---
+
+static void Encode_SensorStateResponse(benchmark::State &state) {
+  APIBuffer buffer;
+  SensorStateResponse msg;
+  msg.key = 0x12345678;
+  msg.state = 23.5f;
+  msg.missing_state = false;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_SensorStateResponse);
+
+static void CalculateSize_SensorStateResponse(benchmark::State &state) {
+  SensorStateResponse msg;
+  msg.key = 0x12345678;
+  msg.state = 23.5f;
+  msg.missing_state = false;
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_SensorStateResponse);
+
+// Steady state: buffer already allocated from previous iteration
+static void CalcAndEncode_SensorStateResponse(benchmark::State &state) {
+  APIBuffer buffer;
+  SensorStateResponse msg;
+  msg.key = 0x12345678;
+  msg.state = 23.5f;
+  msg.missing_state = false;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_SensorStateResponse);
+
+// Cold path: fresh buffer each iteration (measures heap allocation cost).
+// Inner loop still needed to amortize CodSpeed instrumentation overhead.
+// Each inner iteration creates a fresh buffer, so this measures
+// alloc+calc+encode per item.
+static void CalcAndEncode_SensorStateResponse_Fresh(benchmark::State &state) {
+  SensorStateResponse msg;
+  msg.key = 0x12345678;
+  msg.state = 23.5f;
+  msg.missing_state = false;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      APIBuffer buffer;
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+      benchmark::DoNotOptimize(buffer.data());
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_SensorStateResponse_Fresh);
+
+// --- BinarySensorStateResponse ---
+
+static void Encode_BinarySensorStateResponse(benchmark::State &state) {
+  APIBuffer buffer;
+  BinarySensorStateResponse msg;
+  msg.key = 0xAABBCCDD;
+  msg.state = true;
+  msg.missing_state = false;
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_BinarySensorStateResponse);
+
+// --- HelloResponse (string fields) ---
+
+static void Encode_HelloResponse(benchmark::State &state) {
+  APIBuffer buffer;
+  HelloResponse msg;
+  msg.api_version_major = 1;
+  msg.api_version_minor = 10;
+  msg.server_info = StringRef::from_lit("esphome v2026.3.0");
+  msg.name = StringRef::from_lit("living-room-sensor");
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_HelloResponse);
+
+// --- LightStateResponse (complex multi-field message) ---
+
+static void Encode_LightStateResponse(benchmark::State &state) {
+  APIBuffer buffer;
+  LightStateResponse msg;
+  msg.key = 0x11223344;
+  msg.state = true;
+  msg.brightness = 0.8f;
+  msg.color_mode = enums::COLOR_MODE_RGB_WHITE;
+  msg.color_brightness = 1.0f;
+  msg.red = 1.0f;
+  msg.green = 0.5f;
+  msg.blue = 0.2f;
+  msg.white = 0.0f;
+  msg.color_temperature = 4000.0f;
+  msg.cold_white = 0.0f;
+  msg.warm_white = 0.0f;
+  msg.effect = StringRef::from_lit("rainbow");
+  uint32_t size = msg.calculate_size();
+  buffer.resize(size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_LightStateResponse);
+
+static void CalculateSize_LightStateResponse(benchmark::State &state) {
+  LightStateResponse msg;
+  msg.key = 0x11223344;
+  msg.state = true;
+  msg.brightness = 0.8f;
+  msg.color_mode = enums::COLOR_MODE_RGB_WHITE;
+  msg.color_brightness = 1.0f;
+  msg.red = 1.0f;
+  msg.green = 0.5f;
+  msg.blue = 0.2f;
+  msg.white = 0.0f;
+  msg.color_temperature = 4000.0f;
+  msg.cold_white = 0.0f;
+  msg.warm_white = 0.0f;
+  msg.effect = StringRef::from_lit("rainbow");
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_LightStateResponse);
+
+// --- DeviceInfoResponse (nested submessages: 20 devices + 20 areas) ---
+
+static DeviceInfoResponse make_device_info_response() {
+  DeviceInfoResponse msg;
+  msg.name = StringRef::from_lit("living-room-sensor");
+  msg.mac_address = StringRef::from_lit("AA:BB:CC:DD:EE:FF");
+  msg.esphome_version = StringRef::from_lit("2026.3.0");
+  msg.compilation_time = StringRef::from_lit("Mar 16 2026, 12:00:00");
+  msg.model = StringRef::from_lit("esp32-poe-iso");
+  msg.manufacturer = StringRef::from_lit("Olimex");
+  msg.friendly_name = StringRef::from_lit("Living Room Sensor");
+#ifdef USE_DEVICES
+  for (uint32_t i = 0; i < ESPHOME_DEVICE_COUNT && i < 20; i++) {
+    msg.devices[i].device_id = i + 1;
+    msg.devices[i].name = StringRef::from_lit("device");
+    msg.devices[i].area_id = (i % 20) + 1;
+  }
+#endif
+#ifdef USE_AREAS
+  for (uint32_t i = 0; i < ESPHOME_AREA_COUNT && i < 20; i++) {
+    msg.areas[i].area_id = i + 1;
+    msg.areas[i].name = StringRef::from_lit("area");
+  }
+#endif
+  return msg;
+}
+
+static void CalculateSize_DeviceInfoResponse(benchmark::State &state) {
+  auto msg = make_device_info_response();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_DeviceInfoResponse);
+
+static void Encode_DeviceInfoResponse(benchmark::State &state) {
+  auto msg = make_device_info_response();
+  APIBuffer buffer;
+  uint32_t total_size = msg.calculate_size();
+  buffer.resize(total_size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_DeviceInfoResponse);
+
+// Steady state: buffer already allocated from previous iteration
+static void CalcAndEncode_DeviceInfoResponse(benchmark::State &state) {
+  auto msg = make_device_info_response();
+  APIBuffer buffer;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_DeviceInfoResponse);
+
+// Cold path: fresh buffer each iteration (measures heap allocation cost).
+// Inner loop still needed to amortize CodSpeed instrumentation overhead.
+// Each inner iteration creates a fresh buffer, so this measures
+// alloc+calc+encode per item.
+static void CalcAndEncode_DeviceInfoResponse_Fresh(benchmark::State &state) {
+  auto msg = make_device_info_response();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      APIBuffer buffer;
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+      benchmark::DoNotOptimize(buffer.data());
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_DeviceInfoResponse_Fresh);
+
+// --- BluetoothLERawAdvertisementsResponse (12 adverts, highest-volume BLE message) ---
+
+#ifdef USE_BLUETOOTH_PROXY
+
+static BluetoothLERawAdvertisementsResponse make_ble_raw_advs_12() {
+  static const uint8_t fake_adv_data[] = {
+      0x02, 0x01, 0x06, 0x03, 0x03, 0x9F, 0xFE, 0x17, 0x16, 0x9F, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00,
+      0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  };
+  BluetoothLERawAdvertisementsResponse msg;
+  msg.advertisements_len = 12;
+  for (int i = 0; i < 12; i++) {
+    auto &adv = msg.advertisements[i];
+    adv.address = 0xAABBCCDD0000ULL + i;
+    adv.rssi = -60 - i;
+    adv.address_type = 1;
+    memcpy(adv.data, fake_adv_data, sizeof(fake_adv_data));
+    adv.data_len = sizeof(fake_adv_data);
+  }
+  return msg;
+}
+
+static void CalculateSize_BLERawAdvs12(benchmark::State &state) {
+  auto msg = make_ble_raw_advs_12();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_BLERawAdvs12);
+
+static void Encode_BLERawAdvs12(benchmark::State &state) {
+  auto msg = make_ble_raw_advs_12();
+  APIBuffer buffer;
+  uint32_t total_size = msg.calculate_size();
+  buffer.resize(total_size);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_BLERawAdvs12);
+
+static void CalcAndEncode_BLERawAdvs12(benchmark::State &state) {
+  auto msg = make_ble_raw_advs_12();
+  APIBuffer buffer;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_BLERawAdvs12);
+
+static void CalcAndEncode_BLERawAdvs12_Fresh(benchmark::State &state) {
+  auto msg = make_ble_raw_advs_12();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      APIBuffer buffer;
+      uint32_t size = msg.calculate_size();
+      buffer.resize(size);
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+      benchmark::DoNotOptimize(buffer.data());
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalcAndEncode_BLERawAdvs12_Fresh);
+
+#endif  // USE_BLUETOOTH_PROXY
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_proto_proxy.cpp

@@ -0,0 +1,280 @@
+// Encode/decode microbenchmarks for proxy message families that carry
+// high-volume traffic (Z-Wave, IR/RF, serial). Mirrors the existing
+// BluetoothLERawAdvertisementsResponse benchmarks in bench_proto_encode.cpp.
+
+#include <benchmark/benchmark.h>
+
+#include <cstring>
+
+#include "esphome/components/api/api_pb2.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Encodes `src` into `out`. Caller owns `out` and must keep it alive across
+// the decode loop (decoded messages may store pointers back into its bytes).
+template<typename T> static void encode_into(APIBuffer &out, const T &src) {
+  out.resize(src.calculate_size());
+  ProtoWriteBuffer writer(&out, 0);
+  src.encode(writer);
+}
+
+// --- ZWaveProxyFrame (Z-Wave frame, ~16 bytes payload) ---
+
+#ifdef USE_ZWAVE_PROXY
+
+static const uint8_t kZWaveFrameData[] = {0x01, 0x09, 0x00, 0x13, 0x01, 0x02, 0x00, 0x00,
+                                          0x25, 0x00, 0x05, 0xC4, 0x00, 0x00, 0x00, 0x00};
+
+static void Encode_ZWaveProxyFrame(benchmark::State &state) {
+  ZWaveProxyFrame msg;
+  msg.data = kZWaveFrameData;
+  msg.data_len = sizeof(kZWaveFrameData);
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_ZWaveProxyFrame);
+
+static void Decode_ZWaveProxyFrame(benchmark::State &state) {
+  ZWaveProxyFrame source;
+  source.data = kZWaveFrameData;
+  source.data_len = sizeof(kZWaveFrameData);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ZWaveProxyFrame msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_ZWaveProxyFrame);
+
+static const uint8_t kZWaveRequestData[] = {0xDE, 0xAD, 0xBE, 0xEF};
+
+static void Decode_ZWaveProxyRequest(benchmark::State &state) {
+  ZWaveProxyRequest source;
+  source.type = enums::ZWAVE_PROXY_REQUEST_TYPE_HOME_ID_CHANGE;
+  source.data = kZWaveRequestData;
+  source.data_len = sizeof(kZWaveRequestData);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ZWaveProxyRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_ZWaveProxyRequest);
+
+#endif  // USE_ZWAVE_PROXY
+
+// --- SerialProxyDataReceived encode + SerialProxyWriteRequest decode ---
+//
+// SerialProxyWriteRequest is decode-only (SOURCE_CLIENT) but has the same
+// wire layout as SerialProxyDataReceived, so we encode via the latter and
+// decode as the former.
+
+#ifdef USE_SERIAL_PROXY
+
+static constexpr size_t kSerialPayloadSize = 64;
+static const uint8_t kSerialPayload[kSerialPayloadSize] = {
+    0x55, 0xAA, 0x12, 0x34, 0x56, 0x78, 0x9A, 0xBC, 0xDE, 0xF0, 0x01, 0x23, 0x45, 0x67, 0x89, 0xAB,
+    0xCD, 0xEF, 0x11, 0x22, 0x33, 0x44, 0x55, 0x66, 0x77, 0x88, 0x99, 0xAA, 0xBB, 0xCC, 0xDD, 0xEE,
+    0xFF, 0x00, 0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80, 0x90, 0xA0, 0xB0, 0xC0, 0xD0, 0xE0,
+    0xF0, 0x0F, 0x1F, 0x2F, 0x3F, 0x4F, 0x5F, 0x6F, 0x7F, 0x8F, 0x9F, 0xAF, 0xBF, 0xCF, 0xDF, 0xEF};
+
+static void Encode_SerialProxyDataReceived(benchmark::State &state) {
+  SerialProxyDataReceived msg;
+  msg.instance = 0;
+  msg.set_data(kSerialPayload, kSerialPayloadSize);
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_SerialProxyDataReceived);
+
+static void Decode_SerialProxyWriteRequest(benchmark::State &state) {
+  SerialProxyDataReceived source;
+  source.instance = 0;
+  source.set_data(kSerialPayload, kSerialPayloadSize);
+  APIBuffer encoded;
+  encode_into(encoded, source);
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      SerialProxyWriteRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_SerialProxyWriteRequest);
+
+#endif  // USE_SERIAL_PROXY
+
+// --- InfraredRFReceiveEvent encode (100 sint32 timings) +
+//     InfraredRFTransmitRawTimingsRequest decode (hand-built wire bytes) ---
+
+#if defined(USE_IR_RF) || defined(USE_RADIO_FREQUENCY)
+
+// Mark/space pairs simulating a typical RC-5 / NEC capture (100 timings).
+static std::vector<int32_t> make_ir_timings_100() {
+  std::vector<int32_t> v;
+  v.reserve(100);
+  for (int i = 0; i < 100; i++) {
+    v.push_back((i % 2 == 0) ? 560 : -560);
+  }
+  return v;
+}
+
+static const std::vector<int32_t> &get_ir_timings_100() {
+  static const std::vector<int32_t> timings = make_ir_timings_100();
+  return timings;
+}
+
+static void Encode_InfraredRFReceiveEvent(benchmark::State &state) {
+  InfraredRFReceiveEvent msg;
+  msg.key = 0xDEADBEEF;
+  msg.timings = &get_ir_timings_100();
+  APIBuffer buffer;
+  buffer.resize(msg.calculate_size());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      msg.encode(writer);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_InfraredRFReceiveEvent);
+
+static void CalculateSize_InfraredRFReceiveEvent(benchmark::State &state) {
+  InfraredRFReceiveEvent msg;
+  msg.key = 0xDEADBEEF;
+  msg.timings = &get_ir_timings_100();
+
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += msg.calculate_size();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CalculateSize_InfraredRFReceiveEvent);
+
+// Hand-built wire bytes for InfraredRFTransmitRawTimingsRequest (decode-only,
+// no sister message with identical wire layout).
+//   field 2 (key, fixed32):           tag=0x15, 4 LE bytes
+//   field 3 (carrier_frequency):      tag=0x18, varint
+//   field 4 (repeat_count):           tag=0x20, varint
+//   field 5 (timings, packed sint32): tag=0x2A, length varint, packed payload
+//   field 6 (modulation):             tag=0x30, varint
+static APIBuffer build_infrared_rf_transmit_wire() {
+  uint8_t bytes[256];
+  size_t len = 0;
+
+  auto put_byte = [&](uint8_t b) { bytes[len++] = b; };
+  auto put_varint = [&](uint32_t v) {
+    while (v >= 0x80) {
+      bytes[len++] = static_cast<uint8_t>((v & 0x7F) | 0x80);
+      v >>= 7;
+    }
+    bytes[len++] = static_cast<uint8_t>(v);
+  };
+  auto encode_zigzag = [](int32_t v) -> uint32_t {
+    return (static_cast<uint32_t>(v) << 1) ^ static_cast<uint32_t>(v >> 31);
+  };
+
+  put_byte(0x15);
+  put_byte(0xEF);
+  put_byte(0xBE);
+  put_byte(0xAD);
+  put_byte(0xDE);
+  put_byte(0x18);
+  put_varint(38000);
+  put_byte(0x20);
+  put_varint(2);
+
+  uint8_t packed[200];
+  size_t packed_len = 0;
+  for (int i = 0; i < 100; i++) {
+    int32_t value = (i % 2 == 0) ? 560 : -560;
+    uint32_t zz = encode_zigzag(value);
+    while (zz >= 0x80) {
+      packed[packed_len++] = static_cast<uint8_t>((zz & 0x7F) | 0x80);
+      zz >>= 7;
+    }
+    packed[packed_len++] = static_cast<uint8_t>(zz);
+  }
+  put_byte(0x2A);
+  put_varint(static_cast<uint32_t>(packed_len));
+  std::memcpy(bytes + len, packed, packed_len);
+  len += packed_len;
+  // field 6: modulation = 1 (non-zero so it's actually emitted and exercises
+  // decode_varint for this field, matching the documented layout above).
+  put_byte(0x30);
+  put_varint(1);
+
+  APIBuffer buf;
+  buf.resize(len);
+  std::memcpy(buf.data(), bytes, len);
+  return buf;
+}
+
+static void Decode_InfraredRFTransmitRawTimingsRequest(benchmark::State &state) {
+  auto encoded = build_infrared_rf_transmit_wire();
+  const uint8_t *data = encoded.data();
+  size_t size = encoded.size();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      InfraredRFTransmitRawTimingsRequest msg;
+      msg.decode(data, size);
+      benchmark::DoNotOptimize(msg);
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Decode_InfraredRFTransmitRawTimingsRequest);
+
+#endif  // USE_IR_RF || USE_RADIO_FREQUENCY
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_proto_varint.cpp

@@ -0,0 +1,133 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/api/proto.h"
+#include "esphome/components/api/api_buffer.h"
+
+namespace esphome::api::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+// --- ProtoVarInt::parse() benchmarks ---
+
+static void ProtoVarInt_Parse_SingleByte(benchmark::State &state) {
+  uint8_t buf[] = {0x42};  // value = 66
+
+  for (auto _ : state) {
+    ProtoVarIntResult result{};
+    for (int i = 0; i < kInnerIterations; i++) {
+      result = ProtoVarInt::parse(buf, sizeof(buf));
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ProtoVarInt_Parse_SingleByte);
+
+static void ProtoVarInt_Parse_TwoByte(benchmark::State &state) {
+  uint8_t buf[] = {0x80, 0x01};  // value = 128
+
+  for (auto _ : state) {
+    ProtoVarIntResult result{};
+    for (int i = 0; i < kInnerIterations; i++) {
+      result = ProtoVarInt::parse(buf, sizeof(buf));
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ProtoVarInt_Parse_TwoByte);
+
+static void ProtoVarInt_Parse_FiveByte(benchmark::State &state) {
+  uint8_t buf[] = {0xFF, 0xFF, 0xFF, 0xFF, 0x0F};
+
+  for (auto _ : state) {
+    ProtoVarIntResult result{};
+    for (int i = 0; i < kInnerIterations; i++) {
+      result = ProtoVarInt::parse(buf, sizeof(buf));
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ProtoVarInt_Parse_FiveByte);
+
+// --- Varint encoding benchmarks ---
+
+static void Encode_Varint_Small(benchmark::State &state) {
+  APIBuffer buffer;
+  buffer.resize(16);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      writer.encode_varint_raw(42);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_Varint_Small);
+
+static void Encode_Varint_Large(benchmark::State &state) {
+  APIBuffer buffer;
+  buffer.resize(16);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      writer.encode_varint_raw(300);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_Varint_Large);
+
+static void Encode_Varint_MaxUint32(benchmark::State &state) {
+  APIBuffer buffer;
+  buffer.resize(16);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ProtoWriteBuffer writer(&buffer, 0);
+      writer.encode_varint_raw(0xFFFFFFFF);
+    }
+    benchmark::DoNotOptimize(buffer.data());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Encode_Varint_MaxUint32);
+
+// --- ProtoSize::varint() benchmarks ---
+
+static void ProtoSize_Varint_Small(benchmark::State &state) {
+  // Use varying input to prevent constant folding.
+  // Values 0-127 all take 1 byte but the compiler can't prove that.
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += ProtoSize::varint(static_cast<uint32_t>(i) & 0x7F);
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ProtoSize_Varint_Small);
+
+static void ProtoSize_Varint_Large(benchmark::State &state) {
+  // Use varying input to prevent constant folding.
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += ProtoSize::varint(0xFFFF0000 | static_cast<uint32_t>(i));
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ProtoSize_Varint_Large);
+
+}  // namespace esphome::api::benchmarks

tests/benchmarks/components/api/bench_send_sensor_state.cpp

@@ -0,0 +1,191 @@
+#include "esphome/core/defines.h"
+#if defined(USE_API_PLAINTEXT) && defined(USE_SENSOR)
+
+#include <benchmark/benchmark.h>
+#include <unistd.h>
+
+#include "bench_helpers.h"
+#include "esphome/components/api/api_connection.h"
+#include "esphome/components/api/api_server.h"
+#include "esphome/components/sensor/sensor.h"
+
+namespace esphome::api {
+
+// Friend functions declared in APIConnection for benchmark access.
+void bench_enable_immediate_send(APIConnection *conn) { conn->flags_.should_try_send_immediately = true; }
+void bench_clear_batch(APIConnection *conn) { conn->clear_batch_(); }
+void bench_process_batch(APIConnection *conn) { conn->process_batch_(); }
+
+}  // namespace esphome::api
+
+namespace esphome::api::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Helper to create a TCP loopback connection with an APIConnection.
+// Returns the connection and the read-side fd for draining.
+static std::pair<std::unique_ptr<APIConnection>, int> create_api_connection() {
+  auto [sock, read_fd] = create_tcp_loopback();
+  auto conn = std::make_unique<APIConnection>(std::move(sock), global_api_server);
+  conn->start();
+  return {std::move(conn), read_fd};
+}
+
+// Test subclass to access protected configure_entity_() for benchmark setup.
+class TestSensor : public sensor::Sensor {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+};
+
+// --- send_sensor_state: immediate send path ---
+// Measures: send_message_smart_ → prepare buffer → dispatch_message_ →
+// try_send_sensor_state → fill key/device_id + proto encode → frame write →
+// TCP send. This is the per-client cost when batch_delay=0 and initial states
+// have been sent.
+
+static void SendSensorState_Immediate(benchmark::State &state) {
+  auto [conn, read_fd] = create_api_connection();
+  bench_enable_immediate_send(conn.get());
+  // batch_delay must be 0 for should_send_immediately_ to return true
+  uint16_t saved_delay = global_api_server->get_batch_delay();
+  global_api_server->set_batch_delay(0);
+
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+  sensor.publish_state(23.5f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      conn->send_sensor_state(&sensor);
+    }
+    drain_socket(read_fd);
+    benchmark::DoNotOptimize(conn.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  global_api_server->set_batch_delay(saved_delay);
+  ::close(read_fd);
+}
+BENCHMARK(SendSensorState_Immediate);
+
+// --- send_sensor_state: batch path (cold — first call allocates) ---
+// Measures: send_message_smart_ → schedule_message_ → deferred batch add.
+// Includes one-time vector allocation cost.
+
+static void SendSensorState_Batch_Cold(benchmark::State &state) {
+  auto [conn, read_fd] = create_api_connection();
+
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+  sensor.publish_state(23.5f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      conn->send_sensor_state(&sensor);
+    }
+    benchmark::DoNotOptimize(conn.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(SendSensorState_Batch_Cold);
+
+// --- send_sensor_state: batch path (warm — buffer already allocated) ---
+// Measures steady-state batch cost after the vector has been allocated
+// and cleared at least once. This is the typical path during normal
+// operation after the first batch has been processed.
+
+static void SendSensorState_Batch_Warm(benchmark::State &state) {
+  auto [conn, read_fd] = create_api_connection();
+
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+  sensor.publish_state(23.5f);
+
+  // Warm up: send once to allocate, then clear to keep capacity
+  conn->send_sensor_state(&sensor);
+  bench_clear_batch(conn.get());
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      conn->send_sensor_state(&sensor);
+    }
+    benchmark::DoNotOptimize(conn.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(SendSensorState_Batch_Warm);
+
+// --- process_batch_: single sensor state (encode + frame + write) ---
+// Measures the deferred batch processing path: dispatch_message_ →
+// try_send_sensor_state → fill + proto encode → send_buffer → frame write.
+// This is the cost paid on the next loop() after batching.
+
+static void ProcessBatch_SingleSensor(benchmark::State &state) {
+  auto [conn, read_fd] = create_api_connection();
+
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+  sensor.publish_state(23.5f);
+
+  // Warm up batch vector
+  conn->send_sensor_state(&sensor);
+  bench_process_batch(conn.get());
+  drain_socket(read_fd);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      conn->send_sensor_state(&sensor);
+      bench_process_batch(conn.get());
+    }
+    drain_socket(read_fd);
+    benchmark::DoNotOptimize(conn.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(ProcessBatch_SingleSensor);
+
+// --- process_batch_: 5 different sensors ---
+// Measures batch processing with multiple items queued.
+// This exercises the multi-message path in process_batch_.
+
+static void ProcessBatch_5Sensors(benchmark::State &state) {
+  auto [conn, read_fd] = create_api_connection();
+
+  TestSensor sensors[5];
+  for (int i = 0; i < 5; i++) {
+    char name[20];
+    snprintf(name, sizeof(name), "sensor_%d", i);
+    sensors[i].configure(name);
+    sensors[i].publish_state(23.5f + static_cast<float>(i));
+  }
+
+  // Warm up batch vector
+  for (auto &s : sensors)
+    conn->send_sensor_state(&s);
+  bench_process_batch(conn.get());
+  drain_socket(read_fd);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      for (auto &s : sensors)
+        conn->send_sensor_state(&s);
+      bench_process_batch(conn.get());
+    }
+    drain_socket(read_fd);
+    benchmark::DoNotOptimize(conn.get());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+
+  ::close(read_fd);
+}
+BENCHMARK(ProcessBatch_5Sensors);
+
+}  // namespace esphome::api::benchmarks
+
+#endif  // USE_API_PLAINTEXT && USE_SENSOR

tests/benchmarks/components/api/benchmark.yaml

@@ -0,0 +1,115 @@
+# Components needed for API protobuf benchmarks.
+# Merged into the base config before validation so all
+# dependencies get proper defaults.
+#
+# esphome: sub-keys are merged into the base config.
+esphome:
+  areas:
+    - id: area_1
+      name: "Area 1"
+    - id: area_2
+      name: "Area 2"
+    - id: area_3
+      name: "Area 3"
+    - id: area_4
+      name: "Area 4"
+    - id: area_5
+      name: "Area 5"
+    - id: area_6
+      name: "Area 6"
+    - id: area_7
+      name: "Area 7"
+    - id: area_8
+      name: "Area 8"
+    - id: area_9
+      name: "Area 9"
+    - id: area_10
+      name: "Area 10"
+    - id: area_11
+      name: "Area 11"
+    - id: area_12
+      name: "Area 12"
+    - id: area_13
+      name: "Area 13"
+    - id: area_14
+      name: "Area 14"
+    - id: area_15
+      name: "Area 15"
+    - id: area_16
+      name: "Area 16"
+    - id: area_17
+      name: "Area 17"
+    - id: area_18
+      name: "Area 18"
+    - id: area_19
+      name: "Area 19"
+    - id: area_20
+      name: "Area 20"
+  devices:
+    - id: device_1
+      name: "Device 1"
+      area_id: area_1
+    - id: device_2
+      name: "Device 2"
+      area_id: area_2
+    - id: device_3
+      name: "Device 3"
+      area_id: area_3
+    - id: device_4
+      name: "Device 4"
+      area_id: area_4
+    - id: device_5
+      name: "Device 5"
+      area_id: area_5
+    - id: device_6
+      name: "Device 6"
+      area_id: area_6
+    - id: device_7
+      name: "Device 7"
+      area_id: area_7
+    - id: device_8
+      name: "Device 8"
+      area_id: area_8
+    - id: device_9
+      name: "Device 9"
+      area_id: area_9
+    - id: device_10
+      name: "Device 10"
+      area_id: area_10
+    - id: device_11
+      name: "Device 11"
+      area_id: area_11
+    - id: device_12
+      name: "Device 12"
+      area_id: area_12
+    - id: device_13
+      name: "Device 13"
+      area_id: area_13
+    - id: device_14
+      name: "Device 14"
+      area_id: area_14
+    - id: device_15
+      name: "Device 15"
+      area_id: area_15
+    - id: device_16
+      name: "Device 16"
+      area_id: area_16
+    - id: device_17
+      name: "Device 17"
+      area_id: area_17
+    - id: device_18
+      name: "Device 18"
+      area_id: area_18
+    - id: device_19
+      name: "Device 19"
+      area_id: area_19
+    - id: device_20
+      name: "Device 20"
+      area_id: area_20
+
+api:
+  encryption:
+sensor:
+binary_sensor:
+light:
+switch:

tests/benchmarks/components/binary_sensor/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/binary_sensor/bench_binary_sensor_publish.cpp

@@ -0,0 +1,61 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/binary_sensor/binary_sensor.h"
+
+namespace esphome::binary_sensor::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Benchmark: publish_state with alternating values (forces state change every time)
+static void BinarySensorPublish_Alternating(benchmark::State &state) {
+  BinarySensor sensor;
+
+  // First publish to establish initial state
+  sensor.publish_initial_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(i % 2 == 0);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(BinarySensorPublish_Alternating);
+
+// Benchmark: publish_state with same value (tests dedup fast path)
+static void BinarySensorPublish_NoChange(benchmark::State &state) {
+  BinarySensor sensor;
+
+  sensor.publish_initial_state(true);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(true);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(BinarySensorPublish_NoChange);
+
+// Benchmark: publish_state with a callback registered
+static void BinarySensorPublish_WithCallback(benchmark::State &state) {
+  BinarySensor sensor;
+
+  int callback_count = 0;
+  sensor.add_on_state_callback([&callback_count](bool) { callback_count++; });
+
+  sensor.publish_initial_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(i % 2 == 0);
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(BinarySensorPublish_WithCallback);
+
+}  // namespace esphome::binary_sensor::benchmarks

tests/benchmarks/components/binary_sensor/benchmark.yaml

@@ -0,0 +1 @@
+binary_sensor:

tests/benchmarks/components/button/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/button/bench_button.cpp

@@ -0,0 +1,55 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/button/button.h"
+
+namespace esphome::button::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Button for benchmarking — press_action() is a no-op.
+class BenchButton : public Button {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+ protected:
+  void press_action() override {}
+};
+
+// --- Button::press() ---
+// Measures: ESP_LOGD + press_action() + callback dispatch.
+
+static void ButtonPress(benchmark::State &state) {
+  BenchButton button;
+  button.configure("test_button");
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      button.press();
+    }
+    benchmark::DoNotOptimize(&button);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ButtonPress);
+
+// --- Button::press() with callback ---
+// Measures callback dispatch overhead.
+
+static void ButtonPress_WithCallback(benchmark::State &state) {
+  BenchButton button;
+  button.configure("test_button");
+
+  uint64_t callback_count = 0;
+  button.add_on_press_callback([&callback_count]() { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      button.press();
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ButtonPress_WithCallback);
+
+}  // namespace esphome::button::benchmarks

tests/benchmarks/components/button/benchmark.yaml

@@ -0,0 +1 @@
+button:

tests/benchmarks/components/climate/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/climate/bench_climate.cpp

@@ -0,0 +1,142 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/climate/climate.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Climate for benchmarking — control() is a no-op.
+class BenchClimate : public climate::Climate {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+  climate::ClimateTraits traits() override { return this->traits_; }
+
+  climate::ClimateTraits traits_;
+
+ protected:
+  void control(const climate::ClimateCall & /*call*/) override {}
+};
+
+// Helper to create a typical HVAC climate device for benchmarks.
+// Note: setup() is not called (no preferences backend), so save_state_()
+// is effectively a no-op. This benchmarks the call/validation path, not persistence.
+static void setup_hvac_climate(BenchClimate &climate) {
+  climate.configure("test_climate");
+  climate.traits_.set_supported_modes({
+      climate::CLIMATE_MODE_OFF,
+      climate::CLIMATE_MODE_HEAT_COOL,
+      climate::CLIMATE_MODE_COOL,
+      climate::CLIMATE_MODE_HEAT,
+      climate::CLIMATE_MODE_FAN_ONLY,
+  });
+  climate.traits_.set_supported_fan_modes({
+      climate::CLIMATE_FAN_AUTO,
+      climate::CLIMATE_FAN_LOW,
+      climate::CLIMATE_FAN_MEDIUM,
+      climate::CLIMATE_FAN_HIGH,
+  });
+  climate.traits_.set_supported_swing_modes({
+      climate::CLIMATE_SWING_OFF,
+      climate::CLIMATE_SWING_BOTH,
+      climate::CLIMATE_SWING_VERTICAL,
+      climate::CLIMATE_SWING_HORIZONTAL,
+  });
+  climate.traits_.set_supported_presets({
+      climate::CLIMATE_PRESET_NONE,
+      climate::CLIMATE_PRESET_HOME,
+      climate::CLIMATE_PRESET_AWAY,
+  });
+  climate.traits_.set_visual_min_temperature(16.0f);
+  climate.traits_.set_visual_max_temperature(30.0f);
+  climate.traits_.set_visual_target_temperature_step(0.5f);
+  climate.traits_.set_visual_current_temperature_step(0.1f);
+  climate.traits_.add_feature_flags(climate::CLIMATE_SUPPORTS_CURRENT_TEMPERATURE | climate::CLIMATE_SUPPORTS_ACTION);
+}
+
+// --- Climate::publish_state() with temperature update ---
+// Measures the publish path for a thermostat reporting state —
+// the hot path during HVAC operation.
+
+static void ClimatePublish_State(benchmark::State &state) {
+  BenchClimate climate;
+  setup_hvac_climate(climate);
+  climate.mode = climate::CLIMATE_MODE_HEAT;
+  climate.action = climate::CLIMATE_ACTION_HEATING;
+  climate.target_temperature = 22.0f;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      climate.current_temperature = 20.0f + static_cast<float>(i % 100) / 10.0f;
+      climate.publish_state();
+    }
+    benchmark::DoNotOptimize(climate.current_temperature);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ClimatePublish_State);
+
+// --- Climate::publish_state() with callback ---
+// Measures callback dispatch overhead.
+
+static void ClimatePublish_WithCallback(benchmark::State &state) {
+  BenchClimate climate;
+  setup_hvac_climate(climate);
+  climate.mode = climate::CLIMATE_MODE_HEAT;
+  climate.target_temperature = 22.0f;
+
+  uint64_t callback_count = 0;
+  climate.add_on_state_callback([&callback_count](climate::Climate & /*c*/) { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      climate.current_temperature = 20.0f + static_cast<float>(i % 100) / 10.0f;
+      climate.publish_state();
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ClimatePublish_WithCallback);
+
+// --- ClimateCall::perform() set target temperature ---
+// The most common climate call — adjusting the thermostat setpoint.
+
+static void ClimateCall_SetTemperature(benchmark::State &state) {
+  BenchClimate climate;
+  setup_hvac_climate(climate);
+  climate.mode = climate::CLIMATE_MODE_HEAT;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float temp = 18.0f + static_cast<float>(i % 25) * 0.5f;
+      climate.make_call().set_target_temperature(temp).perform();
+    }
+    benchmark::DoNotOptimize(climate.target_temperature);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ClimateCall_SetTemperature);
+
+// --- ClimateCall::perform() mode change with fan ---
+// Exercises the validation path with multiple fields set.
+
+static void ClimateCall_ModeChange(benchmark::State &state) {
+  BenchClimate climate;
+  setup_hvac_climate(climate);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      auto mode = (i % 2 == 0) ? climate::CLIMATE_MODE_HEAT : climate::CLIMATE_MODE_COOL;
+      auto fan = (i % 2 == 0) ? climate::CLIMATE_FAN_HIGH : climate::CLIMATE_FAN_LOW;
+      climate.make_call().set_mode(mode).set_fan_mode(fan).set_target_temperature(22.0f).perform();
+    }
+    benchmark::DoNotOptimize(climate.mode);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ClimateCall_ModeChange);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/climate/benchmark.yaml

@@ -0,0 +1 @@
+climate:

tests/benchmarks/components/core/__init__.py

@@ -0,0 +1,7 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # core (esphome/core/config.py) must run its to_code during builds
+    # because it bootstraps the fundamental application infrastructure.
+    manifest.enable_codegen()

tests/benchmarks/components/cover/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/cover/bench_cover_publish.cpp

@@ -0,0 +1,107 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/cover/cover.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Cover for benchmarking — control() is a no-op.
+class BenchCover : public cover::Cover {
+ public:
+  cover::CoverTraits get_traits() override { return this->traits_; }
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+  cover::CoverTraits traits_;
+
+ protected:
+  void control(const cover::CoverCall & /*call*/) override {}
+};
+
+// --- Cover::publish_state() with position updates ---
+// Measures the publish path for a garage door reporting position
+// during open/close — the hot path during movement.
+
+static void CoverPublish_Position(benchmark::State &state) {
+  BenchCover cover;
+  cover.configure("test_cover");
+  cover.traits_.set_supports_position(true);
+  cover.traits_.set_supports_tilt(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      cover.position = static_cast<float>(i % 101) / 100.0f;
+      cover.current_operation = (i % 2 == 0) ? cover::COVER_OPERATION_OPENING : cover::COVER_OPERATION_CLOSING;
+      cover.publish_state(false);
+    }
+    benchmark::DoNotOptimize(cover.position);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CoverPublish_Position);
+
+// --- Cover::publish_state() with callback ---
+// Measures callback dispatch overhead.
+
+static void CoverPublish_WithCallback(benchmark::State &state) {
+  BenchCover cover;
+  cover.configure("test_cover");
+  cover.traits_.set_supports_position(true);
+
+  uint64_t callback_count = 0;
+  cover.add_on_state_callback([&callback_count]() { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      cover.position = static_cast<float>(i % 101) / 100.0f;
+      cover.publish_state(false);
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CoverPublish_WithCallback);
+
+// --- CoverCall::perform() open/close cycle ---
+// Measures the full call path: validation + control delegation.
+
+static void CoverCall_OpenClose(benchmark::State &state) {
+  BenchCover cover;
+  cover.configure("test_cover");
+  cover.traits_.set_supports_position(true);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      if (i % 2 == 0) {
+        cover.make_call().set_command_open().perform();
+      } else {
+        cover.make_call().set_command_close().perform();
+      }
+    }
+    benchmark::DoNotOptimize(cover.position);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CoverCall_OpenClose);
+
+// --- CoverCall::perform() set position ---
+// Measures the position-setting call path.
+
+static void CoverCall_SetPosition(benchmark::State &state) {
+  BenchCover cover;
+  cover.configure("test_cover");
+  cover.traits_.set_supports_position(true);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float pos = static_cast<float>(i % 101) / 100.0f;
+      cover.make_call().set_position(pos).perform();
+    }
+    benchmark::DoNotOptimize(cover.position);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CoverCall_SetPosition);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/cover/benchmark.yaml

@@ -0,0 +1 @@
+cover:

tests/benchmarks/components/fan/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/fan/bench_fan.cpp

@@ -0,0 +1,122 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/fan/fan.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Fan for benchmarking — control() is a no-op.
+class BenchFan : public fan::Fan {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+  fan::FanTraits get_traits() override { return this->traits_; }
+
+  fan::FanTraits traits_;
+
+ protected:
+  void control(const fan::FanCall & /*call*/) override {}
+};
+
+// Helper to create a typical fan device for benchmarks.
+// Note: setup() is not called (no preferences backend), so save_state_()
+// is effectively a no-op. This benchmarks the call/validation path, not persistence.
+static void setup_fan(BenchFan &fan) {
+  fan.configure("test_fan");
+  fan.traits_.set_oscillation(true);
+  fan.traits_.set_speed(true);
+  fan.traits_.set_supported_speed_count(6);
+  fan.traits_.set_direction(true);
+  fan.set_restore_mode(fan::FanRestoreMode::NO_RESTORE);
+  fan.traits_.set_supported_preset_modes({
+      "auto",
+      "sleep",
+      "nature",
+      "turbo",
+  });
+}
+
+// --- Fan::publish_state() with speed update ---
+// Measures the publish path for a fan reporting state —
+// the hot path during fan operation.
+
+static void FanPublish_State(benchmark::State &state) {
+  BenchFan fan;
+  setup_fan(fan);
+  fan.state = true;
+  fan.direction = fan::FanDirection::FORWARD;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      fan.speed = (i % 6) + 1;
+      fan.publish_state();
+    }
+    benchmark::DoNotOptimize(fan.speed);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FanPublish_State);
+
+// --- Fan::publish_state() with callback ---
+// Measures callback dispatch overhead.
+
+static void FanPublish_WithCallback(benchmark::State &state) {
+  BenchFan fan;
+  setup_fan(fan);
+  fan.state = true;
+
+  uint64_t callback_count = 0;
+  fan.add_on_state_callback([&callback_count]() { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      fan.speed = (i % 6) + 1;
+      fan.publish_state();
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FanPublish_WithCallback);
+
+// --- FanCall::perform() set speed ---
+// The most common fan call — adjusting the speed level.
+
+static void FanCall_SetSpeed(benchmark::State &state) {
+  BenchFan fan;
+  setup_fan(fan);
+  fan.state = true;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      int speed = (i % 6) + 1;
+      fan.make_call().set_speed(speed).perform();
+    }
+    benchmark::DoNotOptimize(fan.speed);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FanCall_SetSpeed);
+
+// --- FanCall::perform() with multiple fields ---
+// Exercises the validation path with state, speed, oscillation, and direction.
+
+static void FanCall_MultiField(benchmark::State &state) {
+  BenchFan fan;
+  setup_fan(fan);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      auto dir = (i % 2 == 0) ? fan::FanDirection::FORWARD : fan::FanDirection::REVERSE;
+      int speed = (i % 6) + 1;
+      fan.make_call().set_state(true).set_speed(speed).set_oscillating(i % 2 == 0).set_direction(dir).perform();
+    }
+    benchmark::DoNotOptimize(fan.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FanCall_MultiField);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/fan/benchmark.yaml

@@ -0,0 +1 @@
+fan:

tests/benchmarks/components/host/__init__.py

@@ -0,0 +1,7 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # host must run its to_code during builds because it sets up
+    # the host platform target execution environment.
+    manifest.enable_codegen()

tests/benchmarks/components/json/__init__.py

@@ -0,0 +1,7 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # json must run its to_code during benchmark builds because it
+    # adds the ArduinoJson library dependency needed by the API component.
+    manifest.enable_codegen()

tests/benchmarks/components/light/__init__.py

@@ -0,0 +1,28 @@
+import esphome.codegen as cg
+from esphome.components.light import generate_gamma_table
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # Light benchmarks need USE_LIGHT_GAMMA_LUT defined and a gamma table
+    # with external linkage that the benchmark .cpp can reference.
+    manifest.enable_codegen()
+    original_to_code = manifest.to_code
+
+    async def to_code(config):
+        await original_to_code(config)
+        cg.add_define("USE_LIGHT_GAMMA_LUT")
+        # Use the light component's own generate_gamma_table() so the
+        # benchmark stays in sync with any formula changes.
+        forward = generate_gamma_table(2.8)
+        values = ", ".join(f"0x{int(v):04X}" for v in forward)
+        # Use extern-visible (non-static) array so the benchmark .cpp
+        # can reference it via extern declaration.
+        cg.add_global(
+            cg.RawStatement(
+                f"extern const uint16_t bench_gamma_2_8_fwd[256] PROGMEM = {{{values}}};"
+            )
+        )
+
+    to_code.priority = original_to_code.priority
+    manifest.to_code = to_code

tests/benchmarks/components/light/bench_light_call.cpp

@@ -0,0 +1,253 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/light/light_output.h"
+#include "esphome/components/light/light_state.h"
+
+// Gamma 2.8 forward LUT generated by the light component's Python codegen
+// (see tests/benchmarks/components/light/__init__.py which calls generate_gamma_table())
+extern const uint16_t bench_gamma_2_8_fwd[256];
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal LightOutput for benchmarking — no real hardware interaction.
+class BenchLightOutput : public light::LightOutput {
+ public:
+  light::LightTraits get_traits() override { return this->traits_; }
+  void write_state(light::LightState * /*state*/) override {}
+
+  light::LightTraits traits_;
+};
+
+// Test subclass to access protected configure_entity_() for benchmark setup.
+class TestLightState : public light::LightState {
+ public:
+  using LightState::LightState;
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+};
+
+// Helper to create a configured RGBWW light state for benchmarks.
+// Note: setup() is not called (no preferences backend), so save_remote_values_()
+// is effectively a no-op. This benchmarks the call/validation path, not persistence.
+static void setup_rgbww_light(BenchLightOutput &output, TestLightState &light) {
+  output.traits_.set_supported_color_modes({light::ColorMode::RGB_COLD_WARM_WHITE});
+  output.traits_.set_min_mireds(153.0f);
+  output.traits_.set_max_mireds(500.0f);
+  light.configure("test_light");
+  light.set_default_transition_length(0);
+  light.set_gamma_correct(2.8f);
+  light.set_gamma_table(bench_gamma_2_8_fwd);
+  light.set_restore_mode(light::LIGHT_ALWAYS_OFF);
+}
+
+// --- LightCall::perform() with instant RGB color change (Home Assistant API path) ---
+// Measures the full call path: validation, set_immediately_, publish, and save.
+// HA sends color_mode explicitly since API 1.6.
+
+static void LightCall_RGBInstant(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  // Turn on first so subsequent calls are color changes
+  light.make_call().set_state(true).set_brightness(1.0f).set_color_brightness(1.0f).set_transition_length(0).perform();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float v = static_cast<float>(i % 256) / 255.0f;
+      light.make_call()
+          .set_color_mode(light::ColorMode::RGB_COLD_WARM_WHITE)
+          .set_red(v)
+          .set_green(1.0f - v)
+          .set_blue(v * 0.5f)
+          .set_transition_length(0)
+          .perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_RGBInstant);
+
+// --- LightCall::perform() turn on/off cycle (Home Assistant API path) ---
+// HA sends color_mode explicitly since API 1.6, skipping compute_color_mode_().
+
+static void LightCall_ToggleOnOff(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      light.make_call()
+          .set_state(i % 2 == 0)
+          .set_color_mode(light::ColorMode::RGB_COLD_WARM_WHITE)
+          .set_transition_length(0)
+          .perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_ToggleOnOff);
+
+// --- LightCall::perform() turn on/off via MQTT ---
+// MQTT never sends color_mode, so compute_color_mode_() runs every call.
+
+static void LightCall_ToggleOnOff_MQTT(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      light.make_call().set_state(i % 2 == 0).set_transition_length(0).perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_ToggleOnOff_MQTT);
+
+// --- LightCall::perform() with color temperature via MQTT ---
+// Exercises the transform_parameters_() path that converts color_temperature
+// to cold/warm white fractions. MQTT never sends color_mode, so this also
+// hits compute_color_mode_() every call. Modern HA avoids this path entirely
+// by converting color temp to CW/WW client-side.
+
+static void LightCall_ColorTemperature_MQTT(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  light.make_call().set_state(true).set_brightness(1.0f).set_transition_length(0).perform();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      // Sweep through color temperature range
+      float ct = 153.0f + static_cast<float>(i % 348);
+      light.make_call().set_color_temperature(ct).set_transition_length(0).perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_ColorTemperature_MQTT);
+
+// --- LightCall::perform() with 1s transition (Home Assistant API path) ---
+// Exercises start_transition_() which allocates a LightTransformer.
+// This is the default HA path when transition_length > 0.
+
+static void LightCall_Transition(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  light.make_call().set_state(true).set_brightness(1.0f).set_transition_length(0).perform();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float v = static_cast<float>(i % 256) / 255.0f;
+      light.make_call()
+          .set_color_mode(light::ColorMode::RGB_COLD_WARM_WHITE)
+          .set_red(v)
+          .set_green(1.0f - v)
+          .set_blue(v * 0.5f)
+          .set_transition_length(1000)
+          .perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_Transition);
+
+// --- LightCall::perform() with cold/warm white (Home Assistant API path) ---
+// Mirrors what modern HA sends: explicit color_mode with direct cold_white
+// and warm_white values. HA converts color temp to CW/WW client-side for
+// CWWW lights (API >= 1.6), so this is the primary HA path.
+
+static void LightCall_ColdWarmWhite(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  light.make_call().set_state(true).set_brightness(1.0f).set_transition_length(0).perform();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float frac = static_cast<float>(i % 256) / 255.0f;
+      light.make_call()
+          .set_color_mode(light::ColorMode::RGB_COLD_WARM_WHITE)
+          .set_cold_white(1.0f - frac)
+          .set_warm_white(frac)
+          .set_transition_length(0)
+          .perform();
+    }
+    benchmark::DoNotOptimize(light.remote_values);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightCall_ColdWarmWhite);
+
+// --- LightState::publish_state() with a remote values listener ---
+// Measures listener notification overhead.
+
+static void LightPublish_WithListener(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  struct TestListener : public light::LightRemoteValuesListener {
+    void on_light_remote_values_update() override { count_++; }
+    uint64_t count_{0};
+  } listener;
+  light.add_remote_values_listener(&listener);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      light.publish_state();
+    }
+    benchmark::DoNotOptimize(listener.count_);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightPublish_WithListener);
+
+// --- current_values_as_rgbww output conversion with gamma LUT ---
+// Measures the output conversion path that real light drivers call
+// from write_state() to get hardware PWM values, including gamma
+// table lookups via the LUT generated by Python codegen.
+
+static void LightOutput_RGBWW(benchmark::State &state) {
+  BenchLightOutput output;
+  TestLightState light(&output);
+  setup_rgbww_light(output, light);
+
+  light.make_call()
+      .set_state(true)
+      .set_brightness(0.8f)
+      .set_color_brightness(0.6f)
+      .set_red(1.0f)
+      .set_green(0.5f)
+      .set_blue(0.2f)
+      .set_cold_white(0.7f)
+      .set_warm_white(0.3f)
+      .set_transition_length(0)
+      .perform();
+
+  float r, g, b, cw, ww;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      light.current_values_as_rgbww(&r, &g, &b, &cw, &ww);
+    }
+    benchmark::DoNotOptimize(r);
+    benchmark::DoNotOptimize(cw);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LightOutput_RGBWW);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/light/benchmark.yaml

@@ -0,0 +1 @@
+light:

tests/benchmarks/components/logger/__init__.py

@@ -0,0 +1,7 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # logger must run its to_code during builds because it configures
+    # the logging subsystem used by ESP_LOG* macros.
+    manifest.enable_codegen()

tests/benchmarks/components/main.cpp

@@ -0,0 +1,42 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/logger/logger.h"
+
+/*
+This special main.cpp provides the entry point for Google Benchmark.
+It replaces the default ESPHome main with a benchmark runner.
+
+*/
+
+// Auto generated code by esphome
+// ========== AUTO GENERATED INCLUDE BLOCK BEGIN ===========
+// ========== AUTO GENERATED INCLUDE BLOCK END ===========
+
+void original_setup() {
+  // Code-generated App initialization (pre_setup, area/device registration, etc.)
+
+  // ========== AUTO GENERATED CODE BEGIN ===========
+  // =========== AUTO GENERATED CODE END ============
+}
+
+void setup() {
+  // Run auto-generated initialization (App.pre_setup, area/device registration,
+  // looping_components_.init, etc.) so benchmarks that use App work correctly.
+  original_setup();
+
+  // Log functions call global_logger->log_vprintf_() without a null check,
+  // so we must set up a Logger before any test that triggers logging.
+  static esphome::logger::Logger test_logger(0);
+  test_logger.set_log_level(ESPHOME_LOG_LEVEL);
+  test_logger.pre_setup();
+
+  int argc = 1;
+  char arg0[] = "benchmark";
+  char *argv[] = {arg0, nullptr};
+  ::benchmark::Initialize(&argc, argv);
+  ::benchmark::RunSpecifiedBenchmarks();
+  ::benchmark::Shutdown();
+  exit(0);
+}
+
+void loop() {}

tests/benchmarks/components/mdns/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/network/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/number/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/number/bench_number.cpp

@@ -0,0 +1,121 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/number/number.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Number for benchmarking — control() publishes the value back.
+class BenchNumber : public number::Number {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+ protected:
+  void control(float value) override { this->publish_state(value); }
+};
+
+// Helper to create a typical number entity for benchmarks.
+static void setup_number(BenchNumber &number) {
+  number.configure("test_number");
+  number.traits.set_min_value(0.0f);
+  number.traits.set_max_value(100.0f);
+  number.traits.set_step(1.0f);
+  number.traits.set_mode(number::NUMBER_MODE_SLIDER);
+}
+
+// --- Number::publish_state() ---
+// Measures the publish path: set_has_state, store value, callback dispatch.
+
+static void NumberPublish_State(benchmark::State &state) {
+  BenchNumber number;
+  setup_number(number);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      number.publish_state(static_cast<float>(i % 100));
+    }
+    benchmark::DoNotOptimize(number.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(NumberPublish_State);
+
+// --- Number::publish_state() with callback ---
+// Measures callback dispatch overhead.
+
+static void NumberPublish_WithCallback(benchmark::State &state) {
+  BenchNumber number;
+  setup_number(number);
+
+  uint64_t callback_count = 0;
+  number.add_on_state_callback([&callback_count](float) { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      number.publish_state(static_cast<float>(i % 100));
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(NumberPublish_WithCallback);
+
+// --- NumberCall::perform() set value ---
+// The most common number call — setting an absolute value.
+// Exercises: validation against min/max, control() dispatch.
+
+static void NumberCall_SetValue(benchmark::State &state) {
+  BenchNumber number;
+  setup_number(number);
+  number.publish_state(50.0f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      float val = static_cast<float>(i % 100);
+      number.make_call().set_value(val).perform();
+    }
+    benchmark::DoNotOptimize(number.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(NumberCall_SetValue);
+
+// --- NumberCall::perform() increment ---
+// Exercises: state read, step arithmetic, max clamping.
+
+static void NumberCall_Increment(benchmark::State &state) {
+  BenchNumber number;
+  setup_number(number);
+  number.publish_state(0.0f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      number.make_call().number_increment(true).perform();
+    }
+    benchmark::DoNotOptimize(number.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(NumberCall_Increment);
+
+// --- NumberCall::perform() decrement ---
+// Exercises: state read, step arithmetic, min clamping.
+
+static void NumberCall_Decrement(benchmark::State &state) {
+  BenchNumber number;
+  setup_number(number);
+  number.publish_state(100.0f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      number.make_call().number_decrement(true).perform();
+    }
+    benchmark::DoNotOptimize(number.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(NumberCall_Decrement);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/number/benchmark.yaml

@@ -0,0 +1 @@
+number:

tests/benchmarks/components/select/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/select/bench_select.cpp

@@ -0,0 +1,157 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/select/select.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Select for benchmarking — control() publishes directly by index.
+class BenchSelect : public select::Select {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+ protected:
+  void control(size_t index) override { this->publish_state(index); }
+};
+
+// Helper to create a select with the given options.
+static void setup_select(BenchSelect &select, const char *name, std::initializer_list<const char *> options) {
+  select.configure(name);
+  select.traits.set_options(options);
+  select.publish_state(size_t(0));
+}
+
+// --- Select::publish_state(size_t) ---
+// The fast path: publish by index, no string lookup.
+
+static void SelectPublish_ByIndex(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.publish_state(static_cast<size_t>(i % 4));
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectPublish_ByIndex);
+
+// --- Select::publish_state(const char *) ---
+// The string path: requires index_of() lookup via strncmp.
+
+static void SelectPublish_ByString(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  const char *options[] = {"off", "still", "move", "still+move"};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.publish_state(options[i % 4]);
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectPublish_ByString);
+
+// --- Select::publish_state() with callback ---
+// Measures callback dispatch overhead on the index path.
+
+static void SelectPublish_WithCallback(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  uint64_t callback_count = 0;
+  select.add_on_state_callback([&callback_count](size_t) { callback_count++; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.publish_state(static_cast<size_t>(i % 4));
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectPublish_WithCallback);
+
+// --- SelectCall::perform() set by index ---
+// The fast call path — no string matching needed.
+
+static void SelectCall_SetByIndex(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.make_call().set_index(i % 4).perform();
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectCall_SetByIndex);
+
+// --- SelectCall::perform() set by option string ---
+// Exercises the string lookup path through index_of().
+
+static void SelectCall_SetByOption(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  const char *options[] = {"off", "still", "move", "still+move"};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.make_call().set_option(options[i % 4]).perform();
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectCall_SetByOption);
+
+// --- SelectCall::perform() next with cycling ---
+// Exercises the navigation path through active_index_.
+
+static void SelectCall_NextCycle(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(select, "test_select", {"off", "still", "move", "still+move"});
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.make_call().select_next(true).perform();
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectCall_NextCycle);
+
+// --- SelectCall with 10 options (string lookup) ---
+// Worst-case string matching with more options.
+
+static void SelectCall_SetByOption_10Options(benchmark::State &state) {
+  BenchSelect select;
+  setup_select(
+      select, "test_select",
+      {"off", "still", "move", "still+move", "custom1", "custom2", "custom3", "custom4", "custom5", "custom6"});
+
+  // Pick options spread across the list to exercise different search depths
+  const char *picks[] = {"off", "custom3", "custom6", "move"};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      select.make_call().set_option(picks[i % 4]).perform();
+    }
+    benchmark::DoNotOptimize(select.active_index());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SelectCall_SetByOption_10Options);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/select/benchmark.yaml

@@ -0,0 +1 @@
+select:

tests/benchmarks/components/sensor/__init__.py

@@ -0,0 +1,12 @@
+import esphome.codegen as cg
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # Sensor filter benchmarks need USE_SENSOR_FILTER defined.
+    # We use a custom to_code instead of enable_codegen() to avoid
+    # pulling in the full sensor component setup.
+    async def to_code(config):
+        cg.add_define("USE_SENSOR_FILTER")
+
+    manifest.to_code = to_code

tests/benchmarks/components/sensor/bench_sensor_filter.cpp

@@ -0,0 +1,78 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/sensor/sensor.h"
+#include "esphome/components/sensor/filter.h"
+
+namespace esphome::sensor::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// Benchmark: sensor publish through a SlidingWindowMovingAverageFilter (window=5, send_every=1)
+static void SensorFilter_SlidingWindowAvg(benchmark::State &state) {
+  Sensor sensor;
+
+  // Create filter: window_size=5, send_every=1, send_first_at=1
+  auto *filter = new SlidingWindowMovingAverageFilter(5, 1, 1);
+  sensor.add_filter(filter);
+
+  float value = 0.0f;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(value);
+      value += 0.1f;
+      if (value > 1000.0f)
+        value = 0.0f;
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorFilter_SlidingWindowAvg);
+
+// Benchmark: sensor publish through ExponentialMovingAverageFilter
+static void SensorFilter_ExponentialMovingAvg(benchmark::State &state) {
+  Sensor sensor;
+
+  // alpha=0.1, send_every=1, send_first_at=1
+  auto *filter = new ExponentialMovingAverageFilter(0.1f, 1, 1);
+  sensor.add_filter(filter);
+
+  float value = 0.0f;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(value);
+      value += 0.1f;
+      if (value > 1000.0f)
+        value = 0.0f;
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorFilter_ExponentialMovingAvg);
+
+// Benchmark: sensor publish through a chain of 3 filters (offset + multiply + sliding window)
+static void SensorFilter_Chain3(benchmark::State &state) {
+  Sensor sensor;
+
+  sensor.add_filters({
+      new OffsetFilter([]() -> float { return 1.0f; }),
+      new MultiplyFilter([]() -> float { return 2.0f; }),
+      new SlidingWindowMovingAverageFilter(5, 1, 1),
+  });
+
+  float value = 0.0f;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(value);
+      value += 0.1f;
+      if (value > 1000.0f)
+        value = 0.0f;
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorFilter_Chain3);
+
+}  // namespace esphome::sensor::benchmarks

tests/benchmarks/components/sensor/bench_sensor_publish.cpp

@@ -0,0 +1,79 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/sensor/sensor.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+// Test subclass to access protected configure_entity_() for benchmark setup.
+class TestSensor : public sensor::Sensor {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+};
+
+// --- Sensor::publish_state() with no callbacks registered ---
+// Measures baseline publish overhead: state assignment, logging,
+// internal_send_state_to_frontend, ControllerRegistry notification.
+
+static void SensorPublish_NoCallbacks(benchmark::State &state) {
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(static_cast<float>(i));
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorPublish_NoCallbacks);
+
+// --- Sensor::publish_state() with one state callback ---
+// Measures callback dispatch overhead through LazyCallbackManager.
+
+static void SensorPublish_WithCallback(benchmark::State &state) {
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+
+  float callback_value = 0.0f;
+  sensor.add_on_state_callback([&callback_value](float value) { callback_value = value; });
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(static_cast<float>(i));
+    }
+    benchmark::DoNotOptimize(callback_value);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorPublish_WithCallback);
+
+// --- Sensor::publish_state() with the same value every time ---
+// Steady-state pattern: sensor reports an unchanged reading.
+// Sensor doesn't dedup today, so this exercises the same code path
+// as changing values, but tracks the common real-world pattern
+// separately for regression detection.
+
+static void SensorPublish_SameValue(benchmark::State &state) {
+  TestSensor sensor;
+  sensor.configure("test_sensor");
+
+  // Warm up so has_state is already set
+  sensor.publish_state(23.5f);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(23.5f);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SensorPublish_SameValue);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/sensor/benchmark.yaml

@@ -0,0 +1 @@
+sensor:

tests/benchmarks/components/socket/__init__.py

@@ -0,0 +1,7 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    # socket must run its to_code to define USE_SOCKET_IMPL_BSD_SOCKETS
+    # which is needed by the api frame helper benchmarks.
+    manifest.enable_codegen()

tests/benchmarks/components/switch/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/switch/bench_switch.cpp

@@ -0,0 +1,137 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/switch/switch.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// Minimal Switch for benchmarking — write_state() publishes directly.
+class BenchSwitch : public switch_::Switch {
+ public:
+  void configure(const char *name) { this->configure_entity_(name, 0x12345678, 0); }
+
+ protected:
+  void write_state(bool state) override { this->publish_state(state); }
+};
+
+// --- Switch::publish_state() alternating ---
+// Forces state change every call, exercising the full publish path.
+
+static void SwitchPublish_Alternating(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+  sw.publish_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.publish_state(i % 2 == 0);
+    }
+    benchmark::DoNotOptimize(sw.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchPublish_Alternating);
+
+// --- Switch::publish_state() no change ---
+// Tests the deduplication fast path in publish_dedup_.
+
+static void SwitchPublish_NoChange(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+  sw.publish_state(true);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.publish_state(true);
+    }
+    benchmark::DoNotOptimize(sw.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchPublish_NoChange);
+
+// --- Switch::publish_state() with callback ---
+// Measures callback dispatch overhead on state changes.
+
+static void SwitchPublish_WithCallback(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+
+  uint64_t callback_count = 0;
+  sw.add_on_state_callback([&callback_count](bool) { callback_count++; });
+  sw.publish_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.publish_state(i % 2 == 0);
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchPublish_WithCallback);
+
+// --- Switch::turn_on() / turn_off() ---
+// The front-end call path: turn_on → write_state → publish_state.
+
+static void SwitchTurnOn(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+  sw.publish_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.turn_on();
+    }
+    benchmark::DoNotOptimize(sw.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchTurnOn);
+
+// --- Switch::toggle() alternating ---
+// Exercises the toggle path which reads current state to determine target.
+
+static void SwitchToggle(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+  sw.publish_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.toggle();
+    }
+    benchmark::DoNotOptimize(sw.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchToggle);
+
+// --- Switch::publish_state() inverted ---
+// Verifies the inversion path doesn't add significant overhead.
+
+static void SwitchPublish_Inverted(benchmark::State &state) {
+  BenchSwitch sw;
+  sw.configure("test_switch");
+  sw.set_restore_mode(switch_::SWITCH_ALWAYS_OFF);
+  sw.set_inverted(true);
+  sw.publish_state(false);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sw.publish_state(i % 2 == 0);
+    }
+    benchmark::DoNotOptimize(sw.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(SwitchPublish_Inverted);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/components/switch/benchmark.yaml

@@ -0,0 +1 @@
+switch:

tests/benchmarks/components/text_sensor/__init__.py

@@ -0,0 +1,5 @@
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    manifest.enable_codegen()

tests/benchmarks/components/text_sensor/bench_text_sensor.cpp

@@ -0,0 +1,108 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/text_sensor/text_sensor.h"
+
+namespace esphome::text_sensor::benchmarks {
+
+static constexpr int kInnerIterations = 2000;
+
+// --- publish_state(const char *) with short string, value changes each time ---
+// Exercises: memcmp check (mismatch), string assign, callback dispatch.
+
+static void TextSensorPublish_Short_Changing(benchmark::State &state) {
+  TextSensor sensor;
+
+  // Pre-populate with different short strings
+  const char *values[] = {"192.168.1.1", "192.168.1.2", "192.168.1.3", "192.168.1.4"};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(values[i % 4]);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(TextSensorPublish_Short_Changing);
+
+// --- publish_state(const char *) with short string, same value (dedup path) ---
+// Exercises: memcmp check (match), skips string assign.
+
+static void TextSensorPublish_Short_NoChange(benchmark::State &state) {
+  TextSensor sensor;
+  sensor.publish_state("192.168.1.100");
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state("192.168.1.100");
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(TextSensorPublish_Short_NoChange);
+
+// --- publish_state with longer string (firmware version, MAC address) ---
+// Exercises: memcmp on longer strings, string assign with potential realloc.
+
+static void TextSensorPublish_Long_Changing(benchmark::State &state) {
+  TextSensor sensor;
+
+  const char *values[] = {
+      "2025.12.0-dev (Jan 15 2025, 10:30:00)",
+      "2025.12.1-dev (Feb 20 2025, 14:45:00)",
+      "2025.12.2-dev (Mar 10 2025, 08:15:00)",
+      "2025.12.3-dev (Apr  5 2025, 16:00:00)",
+  };
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(values[i % 4]);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(TextSensorPublish_Long_Changing);
+
+// --- publish_state with callback ---
+// Measures callback dispatch overhead for text sensors.
+
+static void TextSensorPublish_WithCallback(benchmark::State &state) {
+  TextSensor sensor;
+
+  uint64_t callback_count = 0;
+  sensor.add_on_state_callback([&callback_count](const std::string &) { callback_count++; });
+
+  const char *values[] = {"192.168.1.1", "192.168.1.2", "192.168.1.3", "192.168.1.4"};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(values[i % 4]);
+    }
+    benchmark::DoNotOptimize(callback_count);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(TextSensorPublish_WithCallback);
+
+// --- publish_state(const char *, size_t) direct ---
+// The lowest-level overload, avoids strlen.
+
+static void TextSensorPublish_WithLen(benchmark::State &state) {
+  TextSensor sensor;
+
+  static constexpr const char *values[] = {"192.168.1.1", "192.168.1.2", "192.168.1.3", "192.168.1.4"};
+  static constexpr size_t lens[] = {11, 11, 11, 11};
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      sensor.publish_state(values[i % 4], lens[i % 4]);
+    }
+    benchmark::DoNotOptimize(sensor.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(TextSensorPublish_WithLen);
+
+}  // namespace esphome::text_sensor::benchmarks

tests/benchmarks/components/text_sensor/benchmark.yaml

@@ -0,0 +1 @@
+text_sensor:

tests/benchmarks/components/time/__init__.py

@@ -0,0 +1,9 @@
+import esphome.codegen as cg
+from tests.testing_helpers import ComponentManifestOverride
+
+
+def override_manifest(manifest: ComponentManifestOverride) -> None:
+    async def to_code(config):
+        cg.add_build_flag("-DUSE_TIME_TIMEZONE")
+
+    manifest.to_code = to_code

tests/benchmarks/core/bench_application_loop.cpp

@@ -0,0 +1,22 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/core/application.h"
+
+namespace esphome::benchmarks {
+
+// Benchmark Application::loop() with no registered components.
+// App is initialized by original_setup() in main.cpp (code-generated
+// pre_setup, area/device registration, looping_components_.init).
+// This measures the baseline overhead of the main loop: scheduler,
+// timing, before/after loop tasks, and yield_with_select_.
+static void ApplicationLoop_Empty(benchmark::State &state) {
+  // Set loop interval to 0 so yield_with_select_ returns immediately
+  // instead of sleeping. This benchmarks the loop overhead, not the sleep.
+  App.set_loop_interval(0);
+  for (auto _ : state) {
+    App.loop();
+  }
+}
+BENCHMARK(ApplicationLoop_Empty);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/core/bench_helpers.cpp

@@ -0,0 +1,366 @@
+#include <benchmark/benchmark.h>
+#include <cinttypes>
+#include <cstdio>
+
+#include "esphome/core/helpers.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+// --- random_float() ---
+
+static void RandomFloat(benchmark::State &state) {
+  for (auto _ : state) {
+    float result = 0.0f;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += random_float();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(RandomFloat);
+
+// --- random_uint32() ---
+
+static void RandomUint32(benchmark::State &state) {
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result += random_uint32();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(RandomUint32);
+
+// --- format_hex_to() - 6 bytes (MAC address sized) ---
+
+static void FormatHexTo_6Bytes(benchmark::State &state) {
+  const uint8_t data[] = {0xAB, 0xCD, 0xEF, 0x01, 0x23, 0x45};
+  char buffer[13];  // 6 * 2 + 1
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      format_hex_to(buffer, data, 6);
+    }
+    benchmark::DoNotOptimize(buffer);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FormatHexTo_6Bytes);
+
+// --- format_hex_to() - 16 bytes (UUID sized) ---
+
+static void FormatHexTo_16Bytes(benchmark::State &state) {
+  const uint8_t data[] = {0xAB, 0xCD, 0xEF, 0x01, 0x23, 0x45, 0x67, 0x89,
+                          0xFE, 0xDC, 0xBA, 0x98, 0x76, 0x54, 0x32, 0x10};
+  char buffer[33];  // 16 * 2 + 1
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      format_hex_to(buffer, data, 16);
+    }
+    benchmark::DoNotOptimize(buffer);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FormatHexTo_16Bytes);
+
+// --- format_hex_to() - 100 bytes (large payload) ---
+
+static void FormatHexTo_100Bytes(benchmark::State &state) {
+  uint8_t data[100];
+  for (int i = 0; i < 100; i++) {
+    data[i] = static_cast<uint8_t>(i);
+  }
+  char buffer[201];  // 100 * 2 + 1
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      format_hex_to(buffer, data, 100);
+    }
+    benchmark::DoNotOptimize(buffer);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FormatHexTo_100Bytes);
+
+// --- format_hex_pretty_to() - 6 bytes with ':' separator ---
+
+static void FormatHexPrettyTo_6Bytes(benchmark::State &state) {
+  const uint8_t data[] = {0xAB, 0xCD, 0xEF, 0x01, 0x23, 0x45};
+  char buffer[18];  // 6 * 3
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      format_hex_pretty_to(buffer, data, 6);
+    }
+    benchmark::DoNotOptimize(buffer);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FormatHexPrettyTo_6Bytes);
+
+// --- format_mac_addr_upper() ---
+
+static void FormatMacAddrUpper(benchmark::State &state) {
+  const uint8_t mac[] = {0xAA, 0xBB, 0xCC, 0xDD, 0xEE, 0xFF};
+  char buffer[MAC_ADDRESS_PRETTY_BUFFER_SIZE];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      format_mac_addr_upper(mac, buffer);
+    }
+    benchmark::DoNotOptimize(buffer);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FormatMacAddrUpper);
+
+// --- fnv1_hash() - short string ---
+
+static void Fnv1Hash_Short(benchmark::State &state) {
+  const char *str = "sensor.temperature";
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= fnv1_hash(str);
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Fnv1Hash_Short);
+
+// --- fnv1_hash() - long string ---
+
+static void Fnv1Hash_Long(benchmark::State &state) {
+  const char *str = "binary_sensor.living_room_motion_sensor_occupancy_detected";
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= fnv1_hash(str);
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Fnv1Hash_Long);
+
+// --- fnv1a_hash() - short string ---
+// Use DoNotOptimize on the input pointer to prevent constexpr evaluation
+
+static void Fnv1aHash_Short(benchmark::State &state) {
+  const char *str = "sensor.temperature";
+  benchmark::DoNotOptimize(str);
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= fnv1a_hash(str);
+      benchmark::ClobberMemory();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Fnv1aHash_Short);
+
+// --- fnv1a_hash() - long string ---
+
+static void Fnv1aHash_Long(benchmark::State &state) {
+  const char *str = "binary_sensor.living_room_motion_sensor_occupancy_detected";
+  benchmark::DoNotOptimize(str);
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= fnv1a_hash(str);
+      benchmark::ClobberMemory();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Fnv1aHash_Long);
+
+// --- fnv1_hash_object_id() - typical entity name ---
+
+static void Fnv1HashObjectId(benchmark::State &state) {
+  char name[] = "Living Room Temperature Sensor";
+  size_t len = sizeof(name) - 1;
+  benchmark::DoNotOptimize(name);
+  for (auto _ : state) {
+    uint32_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= fnv1_hash_object_id(name, len);
+      benchmark::ClobberMemory();
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Fnv1HashObjectId);
+
+// --- parse_hex() - 6 bytes from string ---
+
+static void ParseHex_6Bytes(benchmark::State &state) {
+  const char *hex_str = "ABCDEF012345";
+  uint8_t data[6];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      parse_hex(hex_str, data, 6);
+    }
+    benchmark::DoNotOptimize(data);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ParseHex_6Bytes);
+
+// --- parse_hex() - 16 bytes from string ---
+
+static void ParseHex_16Bytes(benchmark::State &state) {
+  const char *hex_str = "ABCDEF0123456789FEDCBA9876543210";
+  uint8_t data[16];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      parse_hex(hex_str, data, 16);
+    }
+    benchmark::DoNotOptimize(data);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ParseHex_16Bytes);
+
+// --- crc8() - 8 bytes ---
+
+static void CRC8_8Bytes(benchmark::State &state) {
+  const uint8_t data[] = {0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08};
+  for (auto _ : state) {
+    uint8_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= crc8(data, 8);
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CRC8_8Bytes);
+
+// --- crc16() - 8 bytes ---
+
+static void CRC16_8Bytes(benchmark::State &state) {
+  const uint8_t data[] = {0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08};
+  for (auto _ : state) {
+    uint16_t result = 0;
+    for (int i = 0; i < kInnerIterations; i++) {
+      result ^= crc16(data, 8);
+    }
+    benchmark::DoNotOptimize(result);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(CRC16_8Bytes);
+
+// --- value_accuracy_to_buf() - typical sensor value ---
+
+static void ValueAccuracyToBuf(benchmark::State &state) {
+  char raw_buf[VALUE_ACCURACY_MAX_LEN] = {};
+  std::span<char, VALUE_ACCURACY_MAX_LEN> buf(raw_buf);
+  float value = 23.456f;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      value_accuracy_to_buf(buf, value, 2);
+    }
+    benchmark::DoNotOptimize(raw_buf);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(ValueAccuracyToBuf);
+
+// --- int8_to_str() ---
+
+static void Int8ToStr(benchmark::State &state) {
+  char buffer[5] = {};
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      int8_to_str(buffer, static_cast<int8_t>(i & 0xFF));
+      benchmark::DoNotOptimize(buffer);
+      benchmark::ClobberMemory();
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Int8ToStr);
+
+// --- base64_decode() - into pre-allocated buffer ---
+
+static void Base64Decode_32Bytes(benchmark::State &state) {
+  // 32 bytes encoded = 44 base64 chars
+  const uint8_t encoded[] = "AAECAwQFBgcICQoLDA0ODxAREhMUFRYXGBkaGx0eHw==";
+  size_t encoded_len = 44;
+  uint8_t output[32];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      base64_decode(encoded, encoded_len, output, sizeof(output));
+    }
+    benchmark::DoNotOptimize(output);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Base64Decode_32Bytes);
+
+// --- uint32_to_str() vs snprintf ---
+
+static void Uint32ToStr_Small(benchmark::State &state) {
+  char buf[UINT32_MAX_STR_SIZE];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_to_str(buf, 12345);
+      benchmark::DoNotOptimize(buf);
+      benchmark::ClobberMemory();
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Uint32ToStr_Small);
+
+static void Snprintf_Uint32_Small(benchmark::State &state) {
+  char buf[UINT32_MAX_STR_SIZE];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      snprintf(buf, sizeof(buf), "%" PRIu32, static_cast<uint32_t>(12345));
+      benchmark::DoNotOptimize(buf);
+      benchmark::ClobberMemory();
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Snprintf_Uint32_Small);
+
+static void Uint32ToStr_Large(benchmark::State &state) {
+  char buf[UINT32_MAX_STR_SIZE];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uint32_to_str(buf, 4294967295u);
+      benchmark::DoNotOptimize(buf);
+      benchmark::ClobberMemory();
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Uint32ToStr_Large);
+
+static void Snprintf_Uint32_Large(benchmark::State &state) {
+  char buf[UINT32_MAX_STR_SIZE];
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      snprintf(buf, sizeof(buf), "%" PRIu32, static_cast<uint32_t>(4294967295u));
+      benchmark::DoNotOptimize(buf);
+      benchmark::ClobberMemory();
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Snprintf_Uint32_Large);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/core/bench_logger.cpp

@@ -0,0 +1,54 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/core/log.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+static constexpr int kInnerIterations = 2000;
+
+static const char *const TAG = "bench";
+
+// --- Log a message with no format specifiers (fastest path) ---
+
+static void Logger_NoFormat(benchmark::State &state) {
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ESP_LOGW(TAG, "Something happened");
+    }
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Logger_NoFormat);
+
+// --- Log a message with 3 uint32_t format specifiers ---
+
+static void Logger_3Uint32(benchmark::State &state) {
+  uint32_t a = 12345, b = 67890, c = 99999;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ESP_LOGW(TAG, "Values: %" PRIu32 " %" PRIu32 " %" PRIu32, a, b, c);
+    }
+    benchmark::DoNotOptimize(a);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Logger_3Uint32);
+
+// --- Log a message with 3 floats (common for sensor values) ---
+
+static void Logger_3Float(benchmark::State &state) {
+  float temp = 23.456f, humidity = 67.89f, pressure = 1013.25f;
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      ESP_LOGW(TAG, "Sensor: %.2f %.1f %.2f", temp, humidity, pressure);
+    }
+    benchmark::DoNotOptimize(temp);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Logger_3Float);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/core/bench_scheduler.cpp

@@ -0,0 +1,233 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/core/scheduler.h"
+#include "esphome/core/hal.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+// Without this, the ~60ns per-iteration valgrind start/stop cost dominates
+// sub-microsecond benchmarks.
+// Must be divisible by all batch sizes used below (3, 10) to avoid
+// pool imbalance at iteration boundaries that causes spurious malloc.
+static constexpr int kInnerIterations = 2100;
+
+// Warm the scheduler pool by registering and replacing items twice.
+// The first batch allocates fresh items; the second batch cancels them and
+// populates the recycling pool with the cancelled items from the first batch.
+static void warm_pool(Scheduler &scheduler, Component *component, int batch_size, uint32_t delay) {
+  uint32_t now = millis();
+  for (int i = 0; i < batch_size; i++) {
+    scheduler.set_timeout(component, static_cast<uint32_t>(i), delay, []() {});
+  }
+  scheduler.call(++now);
+  for (int i = 0; i < batch_size; i++) {
+    scheduler.set_timeout(component, static_cast<uint32_t>(i), delay, []() {});
+  }
+  scheduler.call(++now);
+}
+
+// --- Scheduler fast path: no work to do ---
+
+static void Scheduler_Call_NoWork(benchmark::State &state) {
+  Scheduler scheduler;
+  uint32_t now = millis();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.call(now);
+    }
+    benchmark::DoNotOptimize(now);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_Call_NoWork);
+
+// --- Scheduler with timers: call() when timers exist but aren't due ---
+
+static void Scheduler_Call_TimersNotDue(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // Add some timeouts far in the future
+  for (int i = 0; i < 10; i++) {
+    scheduler.set_timeout(&dummy_component, static_cast<uint32_t>(i), 1000000, []() {});
+  }
+  scheduler.process_to_add();
+
+  uint32_t now = millis();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.call(now);
+    }
+    benchmark::DoNotOptimize(now);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_Call_TimersNotDue);
+
+// --- Scheduler with 5 intervals firing every call ---
+
+static void Scheduler_Call_5IntervalsFiring(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+  int fire_count = 0;
+
+  // Benchmarks the heap-based scheduler dispatch with 5 callbacks firing.
+  // Uses monotonically increasing fake time so intervals reliably fire every call.
+  // USE_BENCHMARK ifdef in component.h disables WarnIfComponentBlockingGuard
+  // (fake now > real millis() would cause underflow in finish()).
+  // interval=0 would cause an infinite loop (reschedules at same now).
+  for (int i = 0; i < 5; i++) {
+    scheduler.set_interval(&dummy_component, static_cast<uint32_t>(i), 1, [&fire_count]() { fire_count++; });
+  }
+  scheduler.process_to_add();
+
+  uint32_t now = millis() + 100;
+
+  for (auto _ : state) {
+    scheduler.call(now);
+    now++;
+    benchmark::DoNotOptimize(fire_count);
+  }
+}
+BENCHMARK(Scheduler_Call_5IntervalsFiring);
+
+// --- Scheduler: set_timeout registration ---
+
+static void Scheduler_SetTimeout(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // Register 3 timeouts then call() — realistic worst case where multiple
+  // components schedule in the same loop iteration. warm_pool fills the
+  // freelist so acquire/recycle never falls back to malloc.
+  static constexpr int kBatchSize = 3;
+  static_assert(kInnerIterations % kBatchSize == 0, "kInnerIterations must be divisible by kBatchSize");
+  warm_pool(scheduler, &dummy_component, kBatchSize, 1000);
+  for (auto _ : state) {
+    uint32_t now = millis();
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.set_timeout(&dummy_component, static_cast<uint32_t>(i % kBatchSize), 1000, []() {});
+      if ((i + 1) % kBatchSize == 0) {
+        scheduler.call(++now);
+      }
+    }
+    scheduler.call(++now);
+    benchmark::DoNotOptimize(scheduler);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_SetTimeout);
+
+// --- Scheduler: set_interval registration ---
+
+static void Scheduler_SetInterval(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // Register 3 intervals then call() — realistic worst case where multiple
+  // components schedule in the same loop iteration. Keeps item count within
+  // the recycling pool (MAX_POOL_SIZE=5) to avoid spurious malloc/free.
+  static constexpr int kBatchSize = 3;
+  static_assert(kInnerIterations % kBatchSize == 0, "kInnerIterations must be divisible by kBatchSize");
+  warm_pool(scheduler, &dummy_component, kBatchSize, 1000);
+  for (auto _ : state) {
+    uint32_t now = millis();
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.set_interval(&dummy_component, static_cast<uint32_t>(i % kBatchSize), 1000, []() {});
+      if ((i + 1) % kBatchSize == 0) {
+        scheduler.call(++now);
+      }
+    }
+    scheduler.call(++now);
+    benchmark::DoNotOptimize(scheduler);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_SetInterval);
+
+// --- Scheduler: defer registration (set_timeout with delay=0) ---
+
+static void Scheduler_Defer(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // defer() is Component::defer which calls set_timeout(delay=0).
+  // Component::defer(func) passes nullptr as the name, which skips
+  // cancel_item_locked_ entirely — matching production behavior where
+  // defers are anonymous fire-and-forget callbacks.
+  static constexpr int kBatchSize = 3;
+  static_assert(kInnerIterations % kBatchSize == 0, "kInnerIterations must be divisible by kBatchSize");
+  warm_pool(scheduler, &dummy_component, kBatchSize, 0);
+  for (auto _ : state) {
+    uint32_t now = millis();
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.set_timeout(&dummy_component, static_cast<const char *>(nullptr), 0, []() {});
+      if ((i + 1) % kBatchSize == 0) {
+        scheduler.call(++now);
+      }
+    }
+    scheduler.call(++now);
+    benchmark::DoNotOptimize(scheduler);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_Defer);
+
+// --- Scheduler: defer with same ID (cancel-and-replace pattern) ---
+
+static void Scheduler_Defer_SameID(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // Measures defer with a fixed numeric ID — each call cancels the previous
+  // pending defer before adding the new one. This is the pattern used by
+  // components that defer work but want to coalesce rapid updates.
+  static constexpr int kBatchSize = 3;
+  static_assert(kInnerIterations % kBatchSize == 0, "kInnerIterations must be divisible by kBatchSize");
+  warm_pool(scheduler, &dummy_component, kBatchSize, 0);
+  for (auto _ : state) {
+    uint32_t now = millis();
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.set_timeout(&dummy_component, static_cast<uint32_t>(0), 0, []() {});
+      if ((i + 1) % kBatchSize == 0) {
+        scheduler.call(++now);
+      }
+    }
+    scheduler.call(++now);
+    benchmark::DoNotOptimize(scheduler);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_Defer_SameID);
+
+// --- Scheduler: set_timeout with batch size exceeding pool (cliff test) ---
+
+static void Scheduler_SetTimeout_ExceedPool(benchmark::State &state) {
+  Scheduler scheduler;
+  Component dummy_component;
+
+  // Register 10 timeouts then call() — larger working set than the 3-item
+  // batches above. With the unbounded freelist, warm_pool preallocates 10
+  // items so this measures steady-state, not malloc cliff.
+  static constexpr int kBatchSize = 10;
+  static_assert(kInnerIterations % kBatchSize == 0, "kInnerIterations must be divisible by kBatchSize");
+  warm_pool(scheduler, &dummy_component, kBatchSize, 1000);
+  for (auto _ : state) {
+    uint32_t now = millis();
+    for (int i = 0; i < kInnerIterations; i++) {
+      scheduler.set_timeout(&dummy_component, static_cast<uint32_t>(i % kBatchSize), 1000, []() {});
+      if ((i + 1) % kBatchSize == 0) {
+        scheduler.call(++now);
+      }
+    }
+    scheduler.call(++now);
+    benchmark::DoNotOptimize(scheduler);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Scheduler_SetTimeout_ExceedPool);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/python/conftest.py

@@ -0,0 +1,22 @@
+"""Shared fixtures for the Python benchmark suite."""
+
+from __future__ import annotations
+
+from collections.abc import Generator
+
+import pytest
+
+from esphome.core import CORE
+
+
+@pytest.fixture(autouse=True)
+def reset_core_state() -> Generator[None]:
+    """Reset CORE before and after every benchmark.
+
+    Per-iteration setups inside benchmarks reset CORE for the loop body;
+    this fixture handles the test-level boundary so stale state from
+    fixture priming doesn't leak across benchmarks.
+    """
+    CORE.reset()
+    yield
+    CORE.reset()

tests/benchmarks/python/fixtures/bluetooth_proxy_device.yaml

@@ -0,0 +1,62 @@
+substitutions:
+  devicename: bluetooth_proxy_device
+  friendly_name: bluetooth_proxy_device
+
+esphome:
+  name: $devicename
+  friendly_name: $friendly_name
+
+esp32:
+  board: esp32-poe-iso
+  framework:
+    type: esp-idf
+    advanced:
+      sram1_as_iram: true
+      minimum_chip_revision: "3.0"
+
+esp32_ble_tracker:
+  scan_parameters:
+    active: false
+
+bluetooth_proxy:
+  active: true
+
+ethernet:
+  type: LAN8720
+  mdc_pin: GPIO23
+  mdio_pin: GPIO18
+  clk_mode: GPIO17_OUT
+  phy_addr: 0
+  power_pin: GPIO12
+
+debug:
+logger:
+api:
+ota:
+  platform: esphome
+
+button:
+  - platform: restart
+    name: Restart
+
+time:
+  - platform: homeassistant
+    id: homeassistant_time
+  - platform: sntp
+    id: sntp_time
+
+sensor:
+  - platform: uptime
+    name: Ethernet Uptime
+  - platform: template
+    name: Free Memory
+    lambda: return heap_caps_get_free_size(MALLOC_CAP_INTERNAL);
+    unit_of_measurement: B
+    state_class: measurement
+  - platform: debug
+    free:
+      name: Heap Free
+    fragmentation:
+      name: Heap Fragmentation
+    min_free:
+      name: Heap Min Free

tests/benchmarks/python/test_compiled_config_bench.py

@@ -0,0 +1,116 @@
+"""CodSpeed benchmarks for the validated-config cache fast path.
+
+PR #16381 added a cache that lets ``esphome upload`` / ``esphome logs``
+skip re-running the full config-validation pipeline. These benchmarks
+compare the cached path (``load_compiled_config``) against the slow
+path (``read_config``) on the same input.
+
+The fixture YAML is a modest bluetooth-proxy device. The two paths
+end up close on a config this small -- the win grows with config
+complexity (external components, large package trees, deeply nested
+schemas), where the slow path can be orders of magnitude slower than
+the cache load.
+
+Skipped when ``pytest-codspeed`` isn't installed so the regular
+unit-test suite keeps working unchanged.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from pathlib import Path
+import shutil
+from typing import Any
+
+import pytest
+
+from esphome.compiled_config import compiled_config_path, load_compiled_config
+from esphome.config import read_config
+from esphome.core import CORE
+from esphome.storage_json import ext_storage_path
+from esphome.writer import update_storage_json
+
+pytest.importorskip("pytest_codspeed")
+
+HERE = Path(__file__).parent
+FIXTURE_YAML = HERE / "fixtures" / "bluetooth_proxy_device.yaml"
+
+
+def _stage_yaml(tmp_path: Path) -> Path:
+    """Copy fixture YAML into a fresh tmp dir.
+
+    Each benchmark gets its own copy so the cache files (under
+    ``.esphome/storage/`` next to the YAML) don't bleed between cases.
+    """
+    target = tmp_path / FIXTURE_YAML.name
+    shutil.copy2(FIXTURE_YAML, target)
+    return target
+
+
+def _prime_cache(yaml_path: Path) -> None:
+    """Run full validation once and persist the cache + sidecar.
+
+    Mirrors ``esphome compile``: ``read_config`` populates ``CORE.config``,
+    then ``update_storage_json`` writes both the StorageJSON sidecar and
+    the ``.validated.yaml`` compiled-config cache.
+    """
+    CORE.config_path = yaml_path
+    config = read_config({}, skip_external_update=True)
+    assert config is not None, f"fixture YAML failed to validate: {yaml_path}"
+    CORE.config = config
+    update_storage_json()
+
+
+@pytest.fixture
+def staged_yaml(tmp_path: Path) -> Path:
+    """YAML copied into tmp_path; no cache files written yet."""
+    return _stage_yaml(tmp_path)
+
+
+@pytest.fixture
+def primed_yaml(staged_yaml: Path) -> Path:
+    """YAML plus a fresh cache + sidecar on disk."""
+    _prime_cache(staged_yaml)
+    assert compiled_config_path(staged_yaml.name).is_file()
+    assert ext_storage_path(staged_yaml.name).is_file()
+    return staged_yaml
+
+
+def _resetting_setup(
+    yaml_path: Path,
+    args: tuple[Any, ...],
+    kwargs: dict[str, Any],
+) -> Callable[[], tuple[tuple[Any, ...], dict[str, Any]]]:
+    """Build a per-iteration setup that resets CORE and re-pins config_path."""
+
+    def setup() -> tuple[tuple[Any, ...], dict[str, Any]]:
+        CORE.reset()
+        CORE.config_path = yaml_path
+        return args, kwargs
+
+    return setup
+
+
+def test_load_compiled_config_cached(primed_yaml: Path, benchmark) -> None:
+    """Fast path: deserialize the cached, already-validated config."""
+    benchmark.pedantic(
+        load_compiled_config,
+        setup=_resetting_setup(primed_yaml, (primed_yaml,), {}),
+        rounds=5,
+        iterations=1,
+    )
+
+
+def test_read_config_uncached(primed_yaml: Path, benchmark) -> None:
+    """Slow path: full validation pipeline (yaml load + schema + components).
+
+    Uses the same primed fixture as the cached path -- ``read_config``
+    ignores the cache file on disk, so the two benchmarks measure the
+    same input from two different code paths.
+    """
+    benchmark.pedantic(
+        read_config,
+        setup=_resetting_setup(primed_yaml, ({},), {"skip_external_update": True}),
+        rounds=3,
+        iterations=1,
+    )

tests/benchmarks/stubs/esphome/components/bluetooth_proxy/bluetooth_proxy.h

@@ -0,0 +1,38 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp needs when USE_BLUETOOTH_PROXY is defined,
+// without pulling in ESP32 BLE dependencies.
+#pragma once
+
+#include "esphome/components/api/api_pb2.h"
+
+namespace esphome {
+namespace api {
+class APIConnection;
+}  // namespace api
+
+namespace bluetooth_proxy {
+
+class BluetoothProxy {
+ public:
+  api::APIConnection *get_api_connection() const { return nullptr; }
+  void subscribe_api_connection(api::APIConnection *conn, uint32_t flags) {}
+  void unsubscribe_api_connection(api::APIConnection *conn) {}
+  void bluetooth_device_request(const api::BluetoothDeviceRequest &msg) {}
+  void bluetooth_gatt_read(const api::BluetoothGATTReadRequest &msg) {}
+  void bluetooth_gatt_write(const api::BluetoothGATTWriteRequest &msg) {}
+  void bluetooth_gatt_read_descriptor(const api::BluetoothGATTReadDescriptorRequest &msg) {}
+  void bluetooth_gatt_write_descriptor(const api::BluetoothGATTWriteDescriptorRequest &msg) {}
+  void bluetooth_gatt_send_services(const api::BluetoothGATTGetServicesRequest &msg) {}
+  void bluetooth_gatt_notify(const api::BluetoothGATTNotifyRequest &msg) {}
+  void send_connections_free(api::APIConnection *conn) {}
+  void bluetooth_scanner_set_mode(bool active) {}
+  void bluetooth_set_connection_params(const api::BluetoothSetConnectionParamsRequest &msg) {}
+  uint32_t get_feature_flags() const { return 0; }
+  void get_bluetooth_mac_address_pretty(char *buf) const { buf[0] = '\0'; }
+};
+
+// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
+extern BluetoothProxy *global_bluetooth_proxy;
+
+}  // namespace bluetooth_proxy
+}  // namespace esphome

tests/benchmarks/stubs/esphome/components/infrared/infrared.h

@@ -0,0 +1,45 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_INFRARED is defined,
+// without pulling in the real remote_base/RMT dependencies.
+#pragma once
+
+#include <cstdint>
+#include "esphome/core/component.h"
+#include "esphome/core/entity_base.h"
+
+namespace esphome::infrared {
+
+class Infrared;
+
+class InfraredCall {
+ public:
+  explicit InfraredCall(Infrared *parent) : parent_(parent) {}
+  InfraredCall &set_carrier_frequency(uint32_t /*frequency*/) { return *this; }
+  InfraredCall &set_raw_timings_packed(const uint8_t * /*data*/, uint16_t /*length*/, uint16_t /*count*/) {
+    return *this;
+  }
+  InfraredCall &set_repeat_count(uint32_t /*count*/) { return *this; }
+  void perform() {}
+
+ protected:
+  Infrared *parent_;
+};
+
+class InfraredTraits {
+ public:
+  uint32_t get_receiver_frequency_hz() const { return 0; }
+};
+
+class Infrared : public Component, public EntityBase {
+ public:
+  Infrared() = default;
+  InfraredTraits &get_traits() { return this->traits_; }
+  const InfraredTraits &get_traits() const { return this->traits_; }
+  InfraredCall make_call() { return InfraredCall(this); }
+  uint32_t get_capability_flags() const { return 0; }
+
+ protected:
+  InfraredTraits traits_;
+};
+
+}  // namespace esphome::infrared

tests/benchmarks/stubs/esphome/components/radio_frequency/radio_frequency.h

@@ -0,0 +1,51 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_RADIO_FREQUENCY is defined.
+#pragma once
+
+#include <cstdint>
+#include "esphome/core/component.h"
+#include "esphome/core/entity_base.h"
+
+namespace esphome::radio_frequency {
+
+enum RadioFrequencyModulation : uint32_t {
+  RADIO_FREQUENCY_MODULATION_OOK = 0,
+};
+
+class RadioFrequency;
+
+class RadioFrequencyCall {
+ public:
+  explicit RadioFrequencyCall(RadioFrequency *parent) : parent_(parent) {}
+  RadioFrequencyCall &set_frequency(uint32_t /*frequency*/) { return *this; }
+  RadioFrequencyCall &set_modulation(RadioFrequencyModulation /*mod*/) { return *this; }
+  RadioFrequencyCall &set_repeat_count(uint32_t /*count*/) { return *this; }
+  RadioFrequencyCall &set_raw_timings_packed(const uint8_t * /*data*/, uint16_t /*length*/, uint16_t /*count*/) {
+    return *this;
+  }
+  void perform() {}
+
+ protected:
+  RadioFrequency *parent_;
+};
+
+class RadioFrequencyTraits {
+ public:
+  uint32_t get_frequency_min_hz() const { return 0; }
+  uint32_t get_frequency_max_hz() const { return 0; }
+  uint32_t get_supported_modulations() const { return 0; }
+};
+
+class RadioFrequency : public Component, public EntityBase {
+ public:
+  RadioFrequency() = default;
+  RadioFrequencyTraits &get_traits() { return this->traits_; }
+  const RadioFrequencyTraits &get_traits() const { return this->traits_; }
+  RadioFrequencyCall make_call() { return RadioFrequencyCall(this); }
+  uint32_t get_capability_flags() const { return 0; }
+
+ protected:
+  RadioFrequencyTraits traits_;
+};
+
+}  // namespace esphome::radio_frequency

tests/benchmarks/stubs/esphome/components/serial_proxy/serial_proxy.h

@@ -0,0 +1,46 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp and Application need when USE_SERIAL_PROXY is defined,
+// without pulling in the real UART implementation.
+#pragma once
+
+#include <cstdint>
+#include <cstddef>
+#include "esphome/components/api/api_pb2.h"
+
+namespace esphome {
+
+namespace api {
+class APIConnection;
+}  // namespace api
+
+namespace uart {
+enum class UARTFlushResult : uint8_t {
+  UART_FLUSH_RESULT_SUCCESS,
+  UART_FLUSH_RESULT_ASSUMED_SUCCESS,
+  UART_FLUSH_RESULT_TIMEOUT,
+  UART_FLUSH_RESULT_FAILED,
+};
+}  // namespace uart
+
+namespace serial_proxy {
+
+class SerialProxy {
+ public:
+  void set_instance_index(uint32_t index) { this->instance_index_ = index; }
+  uint32_t get_instance_index() const { return this->instance_index_; }
+  const char *get_name() const { return ""; }
+  api::enums::SerialProxyPortType get_port_type() const { return {}; }
+  api::APIConnection *get_api_connection() { return nullptr; }
+  void serial_proxy_request(api::APIConnection *conn, api::enums::SerialProxyRequestType type) {}
+  void configure(uint32_t baudrate, bool flow_control, uint8_t parity, uint32_t stop_bits, uint32_t data_size) {}
+  void write_from_client(const uint8_t *data, size_t len) {}
+  void set_modem_pins(uint32_t line_states) {}
+  uint32_t get_modem_pins() const { return 0; }
+  uart::UARTFlushResult flush_port() { return uart::UARTFlushResult::UART_FLUSH_RESULT_SUCCESS; }
+
+ protected:
+  uint32_t instance_index_{0};
+};
+
+}  // namespace serial_proxy
+}  // namespace esphome

tests/benchmarks/stubs/esphome/components/zwave_proxy/zwave_proxy.h

@@ -0,0 +1,29 @@
+// Stub for benchmark builds — provides the minimal interface that
+// api_connection.cpp needs when USE_ZWAVE_PROXY is defined,
+// without pulling in the real UART-based ZWaveProxy implementation.
+#pragma once
+
+#include "esphome/components/api/api_pb2.h"
+
+namespace esphome {
+namespace api {
+class APIConnection;
+}  // namespace api
+
+namespace zwave_proxy {
+
+class ZWaveProxy {
+ public:
+  api::APIConnection *get_api_connection() { return nullptr; }
+  void zwave_proxy_request(api::APIConnection *conn, api::enums::ZWaveProxyRequestType type) {}
+  void send_frame(api::APIConnection *api_connection, const uint8_t *data, size_t length) {}
+  void api_connection_authenticated(api::APIConnection *conn) {}
+  uint32_t get_feature_flags() const { return 0; }
+  uint32_t get_home_id() { return 0; }
+};
+
+// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
+extern ZWaveProxy *global_zwave_proxy;
+
+}  // namespace zwave_proxy
+}  // namespace esphome

tests/component_tests/animation/config/animation_platform_test.yaml

@@ -0,0 +1,33 @@
+# New `image:` `platform: animation` form. Exercises animation/image.py through
+# the real platform loader and codegen pipeline.
+esphome:
+  name: test
+
+esp32:
+  board: esp32s3box
+
+image:
+  - platform: animation
+    id: test_animation
+    file: anim.gif
+    type: rgb565
+    loop:
+      start_frame: 0
+      end_frame: 2
+      repeat: 3
+  - platform: animation
+    id: test_animation_no_loop
+    file: anim.gif
+    type: rgb565
+
+spi:
+  mosi_pin: 6
+  clk_pin: 7
+
+psram:
+  mode: quad
+
+display:
+  - platform: mipi_spi
+    id: lcd_display
+    model: s3box

tests/component_tests/animation/config/animation_test.yaml

@@ -0,0 +1,28 @@
+# Legacy top-level `animation:` form. Exercises the deprecation shim and the
+# shared codegen path through the real read_config/codegen pipeline.
+esphome:
+  name: test
+
+esp32:
+  board: esp32s3box
+
+animation:
+  - id: test_animation
+    file: anim.gif
+    type: rgb565
+    loop:
+      start_frame: 0
+      end_frame: 2
+      repeat: 3
+
+spi:
+  mosi_pin: 6
+  clk_pin: 7
+
+psram:
+  mode: quad
+
+display:
+  - platform: mipi_spi
+    id: lcd_display
+    model: s3box

tests/component_tests/animation/test_init.py

@@ -0,0 +1,81 @@
+"""Tests for the animation image platform and the legacy `animation:` shim."""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+import logging
+from pathlib import Path
+
+import pytest
+
+from esphome.components.animation import (
+    DOMAIN,
+    LEGACY_REMOVAL_VERSION,
+    _capture_legacy_entry,
+    _warn_legacy_animation,
+)
+from esphome.core import CORE
+from esphome.types import ConfigType
+
+# ---------------------------------------------------------------------------
+# Legacy top-level `animation:` deprecation shim -- REMOVE these tests after
+# 2027.1.0 together with the shim in esphome/components/animation/__init__.py.
+# ---------------------------------------------------------------------------
+
+
+def test_warn_legacy_animation_warns_once(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """The deprecation warning fires exactly once and never mutates the config."""
+    config: ConfigType = {"id": "test_animation", "file": "anim.gif", "type": "rgb565"}
+
+    # A per-entry capture (CONFIG_SCHEMA step) records the raw entry so the
+    # one-shot warning can print a pasteable migrated block.
+    assert _capture_legacy_entry(config) is config
+
+    with caplog.at_level(logging.WARNING):
+        # First call: flag not yet set -> warns and records the flag.
+        assert _warn_legacy_animation(config) is config
+        # Second call: flag already set -> stays silent (the dedup branch).
+        assert _warn_legacy_animation(config) is config
+
+    assert CORE.data[DOMAIN]["legacy_warning_shown"] is True
+    warnings = [r for r in caplog.records if r.levelno == logging.WARNING]
+    assert len(warnings) == 1
+    assert "deprecated" in caplog.text
+    assert "platform: animation" in caplog.text
+    assert LEGACY_REMOVAL_VERSION in caplog.text
+
+
+def test_legacy_animation_generation(
+    generate_main: Callable[[str | Path], str],
+    component_config_path: Callable[[str], Path],
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """The legacy `animation:` block validates, warns, and generates codegen
+    through the real read_config/codegen pipeline."""
+    with caplog.at_level(logging.WARNING):
+        main_cpp = generate_main(component_config_path("animation_test.yaml"))
+
+    # Deprecation warning surfaced through the real validation pipeline.
+    assert "animation" in caplog.text
+    assert "deprecated" in caplog.text
+
+    # setup_animation ran: Animation object constructed and loop configured.
+    assert "new(test_animation) animation::Animation(" in main_cpp
+    assert "test_animation->set_loop(0, 2, 3);" in main_cpp
+
+
+def test_animation_platform_generation(
+    generate_main: Callable[[str | Path], str],
+    component_config_path: Callable[[str], Path],
+) -> None:
+    """The `image:` `platform: animation` form generates codegen through the
+    real platform loader (animation/image.py) without any deprecation warning."""
+    main_cpp = generate_main(component_config_path("animation_platform_test.yaml"))
+
+    assert "new(test_animation) animation::Animation(" in main_cpp
+    assert "test_animation->set_loop(0, 2, 3);" in main_cpp
+    # The loop-less entry constructs the object but never configures a loop.
+    assert "new(test_animation_no_loop) animation::Animation(" in main_cpp
+    assert "test_animation_no_loop->set_loop(" not in main_cpp

tests/component_tests/api/test_homeassistant_action.py

@@ -0,0 +1,28 @@
+"""Tests for arg-type selection of api user-defined services with homeassistant.action."""
+
+CONFIG = "tests/component_tests/api/test_homeassistant_action.yaml"
+
+
+def test_synchronous_chain_keeps_zero_copy_args(generate_main):
+    """A chain of synchronous actions keeps the non-owning StringRef arg type."""
+    main_cpp = generate_main(CONFIG)
+
+    assert (
+        "api::UserServiceTrigger<api::enums::SUPPORTS_RESPONSE_NONE, StringRef>"
+        '("zero_copy_args", {"message"})' in main_cpp
+    )
+
+
+def test_response_callback_args_are_owning(generate_main):
+    """homeassistant.action with on_success/on_error stores the trigger args
+    until the HomeassistantActionResponse arrives, so string args must fall
+    back to owning std::string; StringRef would point into the connection's
+    receive buffer, which is reused before the response arrives."""
+    main_cpp = generate_main(CONFIG)
+
+    assert (
+        "api::UserServiceTrigger<api::enums::SUPPORTS_RESPONSE_NONE, std::string>"
+        '("response_args", {"message"})' in main_cpp
+    )
+    assert "api::HomeAssistantServiceCallAction<std::string>" in main_cpp
+    assert "api::HomeAssistantServiceCallAction<StringRef>" not in main_cpp

tests/component_tests/api/test_homeassistant_action.yaml

@@ -0,0 +1,43 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+
+wifi:
+  ssid: SomeNetwork
+  password: SomePassword
+
+logger:
+
+api:
+  actions:
+    # Chain of synchronous actions that never store the args:
+    # keeps the zero-copy StringRef arg type.
+    - action: zero_copy_args
+      variables:
+        message: string
+      then:
+        - logger.log:
+            format: "%s"
+            args: [message.c_str()]
+    # homeassistant.action with on_success/on_error stores the trigger args
+    # until the action response arrives, so the codegen must fall back to
+    # owning std::string args (StringRef would dangle once the receive
+    # buffer is reused).
+    - action: response_args
+      variables:
+        message: string
+      then:
+        - homeassistant.action:
+            action: notify.notify
+            data:
+              message: !lambda return message;
+            on_success:
+              - logger.log:
+                  format: "sent %s"
+                  args: [message.c_str()]
+            on_error:
+              - logger.log:
+                  format: "failed (%s): %s"
+                  args: [error.c_str(), message.c_str()]

tests/component_tests/aqi/test_aqi.py

@@ -0,0 +1,35 @@
+"""Config-validation tests for the aqi sensor component."""
+
+import pytest
+from voluptuous import Invalid
+
+from esphome.components.aqi import CONF_CALCULATION_TYPE, CONF_EXTENDED_RANGE
+from esphome.components.aqi.sensor import _validate_extended_range
+
+
+def test_extended_range_rejected_with_caqi():
+    """extended_range has no meaning for CAQI (no spec maximum) and must be rejected."""
+    with pytest.raises(Invalid, match="CAQI"):
+        _validate_extended_range(
+            {CONF_CALCULATION_TYPE: "CAQI", CONF_EXTENDED_RANGE: True}
+        )
+
+
+def test_extended_range_rejected_with_caqi_even_when_false():
+    """The option is not allowed at all with CAQI, regardless of its value."""
+    with pytest.raises(Invalid, match="CAQI"):
+        _validate_extended_range(
+            {CONF_CALCULATION_TYPE: "CAQI", CONF_EXTENDED_RANGE: False}
+        )
+
+
+def test_extended_range_allowed_with_aqi():
+    """extended_range is valid for the US AQI calculation."""
+    config = {CONF_CALCULATION_TYPE: "AQI", CONF_EXTENDED_RANGE: True}
+    assert _validate_extended_range(config) is config
+
+
+def test_caqi_without_extended_range_ok():
+    """CAQI is fine as long as extended_range is not set."""
+    config = {CONF_CALCULATION_TYPE: "CAQI"}
+    assert _validate_extended_range(config) is config

tests/component_tests/binary_sensor/test_binary_sensor.py

@@ -15,7 +15,7 @@ def test_binary_sensor_is_setup(generate_main):
     )
 
     # Then
-    assert "new gpio::GPIOBinarySensor();" in main_cpp
+    assert "static gpio::GPIOBinarySensor *const" in main_cpp
     assert "App.register_binary_sensor" in main_cpp
 
 
@@ -31,7 +31,7 @@ def test_binary_sensor_sets_mandatory_fields(generate_main):
     )
 
     # Then
-    assert 'bs_1->configure_entity_("test bs1",' in main_cpp
+    assert 'App.register_binary_sensor(bs_1, "test bs1",' in main_cpp
     assert "bs_1->set_pin(" in main_cpp
 
 

tests/component_tests/button/test_button.py

@@ -13,7 +13,8 @@ def test_button_is_setup(generate_main):
     main_cpp = generate_main("tests/component_tests/button/test_button.yaml")
 
     # Then
-    assert "new wake_on_lan::WakeOnLanButton();" in main_cpp
+    assert "static wake_on_lan::WakeOnLanButton *const" in main_cpp
+    assert ") wake_on_lan::WakeOnLanButton();" in main_cpp
     assert "App.register_button" in main_cpp
     assert "App.register_component" in main_cpp
 
@@ -28,7 +29,7 @@ def test_button_sets_mandatory_fields(generate_main):
     main_cpp = generate_main("tests/component_tests/button/test_button.yaml")
 
     # Then
-    assert 'wol_1->configure_entity_("wol_test_1",' in main_cpp
+    assert 'App.register_button(wol_1, "wol_test_1",' in main_cpp
     assert "wol_2->set_macaddr(18, 52, 86, 120, 144, 171);" in main_cpp
 
 

tests/component_tests/conftest.py

@@ -104,6 +104,44 @@ def set_component_config() -> Callable[[str, Any], None]:
     return setter
 
 
+@pytest.fixture
+def choose_variant_with_pins() -> Generator[Callable[[list], None]]:
+    """Set the ESP32 variant to the first one on which all the given pins are valid.
+
+    For ESP32 only, since the other platforms do not have variants. The core
+    configuration must already have been set up for an ESP32 target.
+    Using local imports to avoid importing when ESP32 is not the target.
+    """
+    from esphome import config_validation as cv
+    from esphome.components.esp32 import KEY_ESP32, KEY_VARIANT, VARIANTS
+    from esphome.components.esp32.gpio import validate_gpio_pin
+    from esphome.const import CONF_INPUT, CONF_OUTPUT
+    from esphome.pins import gpio_pin_schema
+
+    def chooser(pins: list) -> None:
+        for variant in VARIANTS:
+            try:
+                CORE.data[KEY_ESP32][KEY_VARIANT] = variant
+                for pin in pins:
+                    if pin is not None:
+                        pin = gpio_pin_schema(
+                            {
+                                CONF_INPUT: True,
+                                CONF_OUTPUT: True,
+                            },
+                            internal=True,
+                        )(pin)
+                        validate_gpio_pin(pin)
+                return
+            except cv.Invalid:
+                continue
+        raise cv.Invalid(
+            f"No compatible variant found for pins: {', '.join(map(str, pins))}"
+        )
+
+    yield chooser
+
+
 @pytest.fixture
 def component_fixture_path(request: pytest.FixtureRequest) -> Callable[[str], Path]:
     """Return a function to get absolute paths relative to the component's fixtures directory."""
@@ -134,7 +172,7 @@ def generate_main() -> Generator[Callable[[str | Path], str]]:
         CORE.config_path = Path(path)
         CORE.config = read_config({})
         generate_cpp_contents(CORE.config)
-        return CORE.cpp_main_section
+        return CORE.cpp_global_section + CORE.cpp_main_section
 
     yield generator
 

tests/component_tests/deep_sleep/test_deep_sleep.py

@@ -7,8 +7,12 @@ def test_deep_sleep_setup(generate_main):
     """
     main_cpp = generate_main("tests/component_tests/deep_sleep/test_deep_sleep1.yaml")
 
-    assert "deepsleep = new deep_sleep::DeepSleepComponent();" in main_cpp
-    assert "App.register_component_(deepsleep);" in main_cpp
+    assert (
+        "static deep_sleep::DeepSleepComponent *const deepsleep = reinterpret_cast<deep_sleep::DeepSleepComponent *>(deep_sleep__deepsleep__pstorage);"
+        in main_cpp
+    )
+    assert "new(deepsleep) deep_sleep::DeepSleepComponent();" in main_cpp
+    assert "App.register_component_(deepsleep, " in main_cpp
 
 
 def test_deep_sleep_sleep_duration(generate_main):
@@ -29,6 +33,43 @@ def test_deep_sleep_run_duration_simple(generate_main):
     assert "deepsleep->set_run_duration(10000);" in main_cpp
 
 
+def test_deep_sleep_on_wake_trigger(generate_main):
+    """
+    When deep sleep is configured with a component-level on_wake automation,
+    a WakeTrigger component should be registered with the wakeup cause as
+    the automation argument.
+    """
+    main_cpp = generate_main("tests/component_tests/deep_sleep/test_deep_sleep3.yaml")
+
+    assert "deep_sleep::WakeTrigger();" in main_cpp
+    assert "Automation<deep_sleep::WakeupCause>" in main_cpp
+
+
+def test_deep_sleep_ext1_on_wake_triggers(generate_main):
+    """
+    Each esp32_ext1_wakeup pin with an on_wake automation should get its own
+    Ext1WakeTrigger with the pin number, and all pins (including the legacy
+    bare-pin shorthand) should contribute to the ext1 wakeup mask.
+    """
+    main_cpp = generate_main("tests/component_tests/deep_sleep/test_deep_sleep3.yaml")
+
+    assert "deep_sleep::Ext1WakeTrigger(2);" in main_cpp
+    assert "deep_sleep::Ext1WakeTrigger(4);" in main_cpp
+    # GPIO13 has no on_wake, so no trigger is created for it
+    assert "deep_sleep::Ext1WakeTrigger(13)" not in main_cpp
+    # mask covers GPIO2, GPIO4 and GPIO13
+    assert ".mask = 8212," in main_cpp
+
+
+def test_deep_sleep_no_on_wake_no_triggers(generate_main):
+    """
+    Without any on_wake automations, no wake trigger code should be generated.
+    """
+    main_cpp = generate_main("tests/component_tests/deep_sleep/test_deep_sleep1.yaml")
+
+    assert "WakeTrigger" not in main_cpp
+
+
 def test_deep_sleep_run_duration_dictionary(generate_main):
     """
     When deep sleep is configured with dictionary run duration, it should be set.

tests/component_tests/deep_sleep/test_deep_sleep3.yaml

@@ -0,0 +1,23 @@
+esphome:
+  name: test
+
+esp32:
+  board: nodemcu-32s
+
+deep_sleep:
+  id: deepsleep
+  sleep_duration: 1min
+  run_duration: 10s
+  on_wake:
+    - lambda: 'ESP_LOGD("test", "cause %d", static_cast<int>(cause));'
+  esp32_ext1_wakeup:
+    mode: ANY_HIGH
+    pins:
+      - pin: GPIO2
+        on_wake:
+          - lambda: 'ESP_LOGD("test", "left");'
+      - pin:
+          number: GPIO4
+        on_wake:
+          - lambda: 'ESP_LOGD("test", "right");'
+      - number: GPIO13

tests/component_tests/display/test_display_metadata.py

@@ -0,0 +1,148 @@
+"""Tests for display component metadata functions."""
+
+from unittest.mock import patch
+
+import pytest
+
+from esphome.components.const import BYTE_ORDER_BIG, BYTE_ORDER_LITTLE
+from esphome.components.display import (
+    DisplayMetaData,
+    add_metadata,
+    get_all_display_metadata,
+    get_display_metadata,
+)
+from esphome.config import Config
+from esphome.core import ID
+from esphome.final_validate import full_config
+
+
+def test_add_metadata_basic():
+    """Test adding metadata with an ID object."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID("my_display"), 320, 240)
+        meta = get_display_metadata(ID("my_display"))
+        assert meta == DisplayMetaData(
+            width=320,
+            height=240,
+            has_hardware_rotation=False,
+            byte_order=BYTE_ORDER_BIG,
+        )
+
+
+def test_add_metadata_with_all_fields():
+    """Test adding metadata with all fields set."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(
+            ID("my_display"),
+            480,
+            320,
+            has_hardware_rotation=True,
+            byte_order=BYTE_ORDER_LITTLE,
+        )
+        meta = get_display_metadata(ID("my_display"))
+        assert meta == DisplayMetaData(
+            width=480,
+            height=320,
+            has_hardware_rotation=True,
+            byte_order=BYTE_ORDER_LITTLE,
+        )
+
+
+def test_add_metadata_hardware_rotation_default():
+    """Test that has_hardware_rotation defaults to False."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID("disp"), 128, 64)
+        meta = get_display_metadata(ID("disp"))
+        assert meta.has_hardware_rotation is False
+        assert meta.byte_order == BYTE_ORDER_BIG
+
+
+def test_add_metadata_with_byte_order():
+    """Test adding metadata with explicit byte_order."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID("disp"), 240, 320, byte_order=BYTE_ORDER_LITTLE)
+        meta = get_display_metadata(ID("disp"))
+        assert meta.byte_order == BYTE_ORDER_LITTLE
+
+
+def test_get_display_metadata_missing_reads_raw_config():
+    """Querying a non-existent ID falls back to raw config lookup."""
+    with patch("esphome.components.display.CORE.data", {}):
+        # Set up a minimal full_config with a display entry so the fallback
+        # path in get_display_metadata can find the display config.
+        fc = Config()
+        fc["display"] = [
+            {
+                "id": ID("no_such_display", True),
+                "auto_clear_enabled": True,
+                "dimensions": {"width": 320, "height": 240},
+                "byte_order": BYTE_ORDER_LITTLE,
+                "rotation": 90,
+            },
+            {
+                "id": ID("other_display", True),
+                "auto_clear_enabled": "undefined",
+                "dimensions": (1024, 600),
+            },
+        ]
+        fc.declare_ids.append((ID("no_such_display", True), ["display", 0, "id"]))
+        fc.declare_ids.append((ID("other_display", True), ["display", 1, "id"]))
+        full_config.set(fc)
+        data = get_display_metadata(ID("no_such_display"))
+        assert data.width == 320
+        assert data.height == 240
+        assert data.has_hardware_rotation is False
+        assert data.has_writer is True
+        assert data.byte_order == BYTE_ORDER_LITTLE
+        assert data.rotation == 90
+
+        data = get_display_metadata(ID("other_display"))
+        assert data.width == 1024
+        assert data.height == 600
+        assert data.has_writer is False
+
+
+def test_add_multiple_displays():
+    """Test adding metadata for multiple displays."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID("disp_a"), 320, 240)
+        add_metadata(ID("disp_b"), 128, 64, has_hardware_rotation=True)
+
+        all_meta = get_all_display_metadata()
+        assert len(all_meta) == 2
+        assert all_meta["disp_a"] == DisplayMetaData(320, 240, False)
+        assert all_meta["disp_b"] == DisplayMetaData(128, 64, True, BYTE_ORDER_BIG)
+
+
+def test_add_duplicate_id_asserts():
+    """Adding metadata for the same ID object twice should assert."""
+    with patch("esphome.components.display.CORE.data", {}):
+        id_obj = ID("disp")
+        add_metadata(id_obj, 320, 240)
+        with pytest.raises(AssertionError, match="Duplicate"):
+            add_metadata(id_obj, 640, 480)
+
+
+def test_metadata_is_frozen():
+    """Test that DisplayMetaData instances are immutable (frozen dataclass)."""
+    meta = DisplayMetaData(320, 240, False, BYTE_ORDER_BIG)
+    with pytest.raises(AttributeError):
+        meta.width = 640
+    with pytest.raises(AttributeError):
+        meta.byte_order = BYTE_ORDER_LITTLE
+
+
+def test_get_all_metadata_asserts_on_unresolved_id():
+    """get_all_display_metadata should assert if any ID has id=None."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID(None), 320, 240)
+        with pytest.raises(AssertionError, match="resolved"):
+            get_all_display_metadata()
+
+
+def test_get_metadata_asserts_on_unresolved_id():
+    """get_display_metadata should assert if any ID has id=None."""
+    with patch("esphome.components.display.CORE.data", {}):
+        add_metadata(ID(None), 320, 240)
+        with pytest.raises(AssertionError, match="resolved"):
+            get_display_metadata(ID("anything"))

tests/component_tests/epaper_spi/config/enable_pin_test.yaml

@@ -0,0 +1,24 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+
+spi:
+  clk_pin: GPIO18
+  mosi_pin: GPIO19
+
+display:
+  - platform: epaper_spi
+    id: epaper_display
+    model: ssd1677
+    dc_pin: GPIO21
+    busy_pin: GPIO22
+    reset_pin: GPIO23
+    cs_pin: GPIO5
+    enable_pin:
+      - GPIO25
+      - GPIO26
+    dimensions:
+      width: 200
+      height: 200

tests/component_tests/epaper_spi/test_display_metadata.py

@@ -0,0 +1,156 @@
+"""Tests for display metadata created by the epaper_spi component."""
+
+from collections.abc import Callable
+from pathlib import Path
+from typing import Any
+
+from esphome import config_validation as cv
+from esphome.components.display import get_all_display_metadata, get_display_metadata
+from esphome.components.epaper_spi.display import CONFIG_SCHEMA
+from esphome.components.esp32 import KEY_BOARD, KEY_VARIANT, VARIANT_ESP32
+from esphome.const import PlatformFramework
+from esphome.types import ConfigType
+from tests.component_tests.types import SetCoreConfigCallable
+
+
+def _base_config(**overrides: Any) -> ConfigType:
+    """Build a minimal valid ssd1677 config, allowing field overrides."""
+    config: ConfigType = {
+        "id": "test_display",
+        "model": "ssd1677",
+        "dc_pin": 21,
+        "busy_pin": 22,
+        "reset_pin": 23,
+        "cs_pin": 5,
+        "dimensions": {"width": 200, "height": 300},
+    }
+    config.update(overrides)
+    return config
+
+
+def test_metadata_dimensions_and_defaults(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """Metadata picks up explicit dimensions and epaper_spi defaults."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    config = CONFIG_SCHEMA(_base_config())
+    meta = get_display_metadata(config["id"])
+
+    assert meta is not None
+    assert meta.width == 200
+    assert meta.height == 300
+    # epaper_spi always reports full hardware rotation
+    assert meta.has_hardware_rotation is True
+    # epaper_spi does not declare a byte order
+    assert meta.byte_order is cv.UNDEFINED
+    assert meta.draw_rounding == 0
+    # no drawing methods configured -> no writer
+    assert meta.has_writer is False
+
+
+def test_metadata_default_dimensions_from_model(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """A model with built-in dimensions reports those without explicit dimensions."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    # waveshare-4.26in is an ssd1677 derivative with default 800x480 dimensions
+    config = CONFIG_SCHEMA(
+        {
+            "id": "wave_display",
+            "model": "waveshare-4.26in",
+            "dc_pin": 21,
+            "busy_pin": 22,
+            "reset_pin": 23,
+            "cs_pin": 5,
+        }
+    )
+    meta = get_display_metadata(config["id"])
+
+    assert meta is not None
+    assert meta.width == 800
+    assert meta.height == 480
+
+
+def test_metadata_has_writer_with_auto_clear(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """A display with auto_clear_enabled reports has_writer=True."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    config = CONFIG_SCHEMA(_base_config(auto_clear_enabled=True))
+    meta = get_display_metadata(config["id"])
+
+    assert meta is not None
+    assert meta.has_writer is True
+
+
+def test_metadata_rotation_propagated(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """The configured rotation is stored in the metadata."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    config = CONFIG_SCHEMA(_base_config(rotation=90))
+    meta = get_display_metadata(config["id"])
+
+    assert meta is not None
+    assert meta.rotation == 90
+
+
+def test_metadata_multiple_displays_independent(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """Each display gets its own independent metadata entry."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    CONFIG_SCHEMA(_base_config(id="disp_a", dimensions={"width": 200, "height": 300}))
+    CONFIG_SCHEMA(_base_config(id="disp_b", dimensions={"width": 400, "height": 480}))
+
+    all_meta = get_all_display_metadata()
+    assert all_meta["disp_a"].width == 200
+    assert all_meta["disp_a"].height == 300
+    assert all_meta["disp_b"].width == 400
+    assert all_meta["disp_b"].height == 480
+
+
+def test_metadata_via_code_generation(
+    generate_main: Callable[[str | Path], str],
+    component_config_path: Callable[[str], Path],
+) -> None:
+    """Full code generation registers metadata for the configured display."""
+    generate_main(component_config_path("enable_pin_test.yaml"))
+
+    all_meta = get_all_display_metadata()
+    assert len(all_meta) == 1
+    meta = next(iter(all_meta.values()))
+    # enable_pin_test.yaml: ssd1677 at 200x200
+    assert meta.width == 200
+    assert meta.height == 200
+    assert meta.has_hardware_rotation is True

tests/component_tests/epaper_spi/test_init.py

@@ -1,6 +1,8 @@
 """Tests for epaper_spi configuration validation."""
 
 from collections.abc import Callable
+from pathlib import Path
+import re
 from typing import Any
 
 import pytest
@@ -11,17 +13,13 @@ from esphome.components.epaper_spi.display import (
     FINAL_VALIDATE_SCHEMA,
     MODELS,
 )
-from esphome.components.esp32 import (
-    KEY_BOARD,
-    KEY_VARIANT,
-    VARIANT_ESP32,
-    VARIANT_ESP32S3,
-)
+from esphome.components.esp32 import KEY_BOARD, KEY_VARIANT, VARIANT_ESP32
 from esphome.const import (
     CONF_BUSY_PIN,
     CONF_CS_PIN,
     CONF_DC_PIN,
     CONF_DIMENSIONS,
+    CONF_ENABLE_PIN,
     CONF_HEIGHT,
     CONF_INIT_SEQUENCE,
     CONF_RESET_PIN,
@@ -31,6 +29,30 @@ from esphome.const import (
 from esphome.types import ConfigType
 from tests.component_tests.types import SetCoreConfigCallable
 
+# Pin options whose values must be valid on the chosen ESP32 variant.
+_PIN_CONF_KEYS = (
+    CONF_CS_PIN,
+    CONF_DC_PIN,
+    CONF_RESET_PIN,
+    CONF_BUSY_PIN,
+    CONF_ENABLE_PIN,
+)
+
+
+def _pins_for(model: Any, config: ConfigType) -> list:
+    """Collect every GPIO the config will actually use (model defaults or injected)."""
+    pins: list = []
+    for key in _PIN_CONF_KEYS:
+        # An injected value in the config takes precedence over the model default.
+        value = config[key] if key in config else model.get_default(key)
+        if not value:  # get_default returns False for pins the model omits
+            continue
+        if isinstance(value, list):
+            pins.extend(value)
+        else:
+            pins.append(value)
+    return pins
+
 
 def run_schema_validation(
     config: ConfigType, with_final_validate: bool = False
@@ -90,29 +112,20 @@ def test_basic_configuration_errors(
 def test_all_predefined_models(
     set_core_config: SetCoreConfigCallable,
     set_component_config: Callable[[str, Any], None],
+    choose_variant_with_pins: Callable[[list], None],
 ) -> None:
     """Test all predefined epaper models validate successfully with appropriate defaults."""
 
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+
+    # Configure SPI component which is required by epaper_spi
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
     # Test all models, providing default values where necessary
     for name, model in MODELS.items():
-        # SEEED models are designed for ESP32-S3 hardware
-        if name in ("SEEED-EE04-MONO-4.26", "SEEED-RETERMINAL-E1002"):
-            set_core_config(
-                PlatformFramework.ESP32_IDF,
-                platform_data={
-                    KEY_BOARD: "esp32-s3-devkitc-1",
-                    KEY_VARIANT: VARIANT_ESP32S3,
-                },
-            )
-        else:
-            set_core_config(
-                PlatformFramework.ESP32_IDF,
-                platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
-            )
-
-        # Configure SPI component which is required by epaper_spi
-        set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
-
         config = {"model": name}
 
         # Add ID field
@@ -141,6 +154,14 @@ def test_all_predefined_models(
         if not model.get_default(CONF_CS_PIN):
             config[CONF_CS_PIN] = 5
 
+        # Dual-CS models (e.g. T133A01) require a second chip-select pin
+        if model.manages_cs and not model.get_default("cs1_pin"):
+            config["cs1_pin"] = 4
+
+        # Select an ESP32 variant on which all of this model's pins are valid
+        # (some models default to high-numbered pins only present on the S3).
+        choose_variant_with_pins(_pins_for(model, config))
+
         run_schema_validation(config)
 
 
@@ -152,27 +173,19 @@ def test_individual_models(
     model_name: str,
     set_core_config: SetCoreConfigCallable,
     set_component_config: Callable[[str, Any], None],
+    choose_variant_with_pins: Callable[[list], None],
 ) -> None:
     """Test each epaper model individually to ensure it validates correctly."""
-    # SEEED models are designed for ESP32-S3 hardware
-    if model_name in ("SEEED-EE04-MONO-4.26", "SEEED-RETERMINAL-E1002"):
-        set_core_config(
-            PlatformFramework.ESP32_IDF,
-            platform_data={
-                KEY_BOARD: "esp32-s3-devkitc-1",
-                KEY_VARIANT: VARIANT_ESP32S3,
-            },
-        )
-    else:
-        set_core_config(
-            PlatformFramework.ESP32_IDF,
-            platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
-        )
+    model = MODELS[model_name]
+
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
 
     # Configure SPI component which is required by epaper_spi
     set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
 
-    model = MODELS[model_name]
     config: dict[str, Any] = {"model": model_name, "id": "test_display"}
 
     # Add required fields based on model defaults
@@ -195,6 +208,14 @@ def test_individual_models(
     if not model.get_default(CONF_CS_PIN):
         config[CONF_CS_PIN] = 5
 
+    # Dual-CS models (e.g. T133A01) require a second chip-select pin
+    if model.manages_cs and not model.get_default("cs1_pin"):
+        config["cs1_pin"] = 4
+
+    # Select an ESP32 variant on which all of this model's pins are valid
+    # (some models default to high-numbered pins only present on the S3).
+    choose_variant_with_pins(_pins_for(model, config))
+
     # This should not raise any exceptions
     run_schema_validation(config)
 
@@ -342,3 +363,102 @@ def test_busy_pin_input_mode_ssd1677(
     reset_pin_config = result[CONF_RESET_PIN]
     assert "mode" in reset_pin_config
     assert reset_pin_config["mode"]["output"] is True
+
+
+def test_enable_pin_single(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """Test that a single enable_pin is accepted and normalised to a list of output pins."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+
+    # Configure SPI component which is required by epaper_spi
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    result = run_schema_validation(
+        {
+            "id": "test_display",
+            "model": "ssd1677",
+            "dc_pin": 21,
+            "busy_pin": 22,
+            "reset_pin": 23,
+            "cs_pin": 5,
+            "enable_pin": 25,
+            "dimensions": {
+                "width": 200,
+                "height": 200,
+            },
+        }
+    )
+
+    # A single pin is normalised to a list by cv.ensure_list
+    assert CONF_ENABLE_PIN in result
+    enable_pins = result[CONF_ENABLE_PIN]
+    assert isinstance(enable_pins, list)
+    assert len(enable_pins) == 1
+    # enable pins are configured as outputs
+    assert enable_pins[0]["mode"]["output"] is True
+
+
+def test_enable_pin_multiple(
+    set_core_config: SetCoreConfigCallable,
+    set_component_config: Callable[[str, Any], None],
+) -> None:
+    """Test that a list of enable_pins is accepted."""
+    set_core_config(
+        PlatformFramework.ESP32_IDF,
+        platform_data={KEY_BOARD: "esp32dev", KEY_VARIANT: VARIANT_ESP32},
+    )
+
+    # Configure SPI component which is required by epaper_spi
+    set_component_config("spi", {"id": "spi_bus", "clk_pin": 18, "mosi_pin": 19})
+
+    result = run_schema_validation(
+        {
+            "id": "test_display",
+            "model": "ssd1677",
+            "dc_pin": 21,
+            "busy_pin": 22,
+            "reset_pin": 23,
+            "cs_pin": 5,
+            "enable_pin": [25, 26],
+            "dimensions": {
+                "width": 200,
+                "height": 200,
+            },
+        }
+    )
+
+    assert CONF_ENABLE_PIN in result
+    enable_pins = result[CONF_ENABLE_PIN]
+    assert isinstance(enable_pins, list)
+    assert len(enable_pins) == 2
+    assert all(pin["mode"]["output"] is True for pin in enable_pins)
+
+
+def test_enable_pin_code_generation(
+    generate_main: Callable[[str | Path], str],
+    component_config_path: Callable[[str], Path],
+) -> None:
+    """Test that enable_pins are wired up in the generated C++ code."""
+    main_cpp = generate_main(component_config_path("enable_pin_test.yaml"))
+
+    # Derive the auto-generated pin variable names from the set_pin() lines
+    # rather than hard-coding them, so the test does not break when unrelated
+    # codegen details shift the generated IDs.
+    def pin_var_for(gpio_num: int) -> str:
+        match = re.search(rf"(\w+)->set_pin\(::GPIO_NUM_{gpio_num}\);", main_cpp)
+        assert match is not None, (
+            f"GPIO_NUM_{gpio_num} pin not set up in generated code"
+        )
+        return match.group(1)
+
+    pin_25 = pin_var_for(25)
+    pin_26 = pin_var_for(26)
+
+    # Both pin objects must be passed to the display via set_enable_pins() as a
+    # std::vector initializer list, in the configured order.
+    assert f"set_enable_pins({{{pin_25}, {pin_26}}});" in main_cpp

tests/component_tests/esp32/config/flash_mode_default.yaml

@@ -0,0 +1,7 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  framework:
+    type: esp-idf

tests/component_tests/esp32/config/flash_mode_idf.yaml

@@ -0,0 +1,10 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  flash_mode: qio
+  flash_frequency: 80MHz
+  toolchain: platformio
+  framework:
+    type: esp-idf

tests/component_tests/esp32/config/network_ethernet_only.yaml

@@ -0,0 +1,17 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  framework:
+    type: esp-idf
+
+ethernet:
+  type: W5500
+  clk_pin: 19
+  mosi_pin: 21
+  miso_pin: 23
+  cs_pin: 18
+  interrupt_pin: 36
+  reset_pin: 22
+  clock_speed: 10Mhz

tests/component_tests/esp32/config/network_wifi_ble_coexistence.yaml

@@ -0,0 +1,14 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  framework:
+    type: esp-idf
+
+wifi:
+  ssid: "test_ssid"
+  password: "test_password"
+
+esp32_ble_tracker:
+  software_coexistence: true

tests/component_tests/esp32/config/network_wifi_ethernet_priority.yaml

@@ -0,0 +1,26 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  framework:
+    type: esp-idf
+
+wifi:
+  ssid: "test_ssid"
+  password: "test_password"
+
+ethernet:
+  type: W5500
+  clk_pin: 19
+  mosi_pin: 21
+  miso_pin: 23
+  cs_pin: 18
+  interrupt_pin: 36
+  reset_pin: 22
+  clock_speed: 10Mhz
+
+network:
+  priority:
+    - ethernet
+    - wifi

tests/component_tests/esp32/config/network_wifi_only.yaml

@@ -0,0 +1,11 @@
+esphome:
+  name: test
+
+esp32:
+  board: esp32dev
+  framework:
+    type: esp-idf
+
+wifi:
+  ssid: "test_ssid"
+  password: "test_password"

tests/component_tests/esp32/config/nvs_encryption_s3.yaml

@@ -0,0 +1,10 @@
+esphome:
+  name: test
+
+esp32:
+  variant: esp32s3
+  framework:
+    type: esp-idf
+    advanced:
+      nvs_encryption:
+        key_id: 0

tests/component_tests/esp32/config/psram_octal_disabled_gpio34.yaml

@@ -0,0 +1,16 @@
+esphome:
+  name: test
+
+esp32:
+  variant: esp32s3
+  framework:
+    type: esp-idf
+
+psram:
+  mode: octal
+  disabled: true
+
+binary_sensor:
+  - platform: gpio
+    pin: GPIO34
+    name: test

tests/component_tests/esp32/config/psram_octal_gpio34.yaml

@@ -0,0 +1,15 @@
+esphome:
+  name: test
+
+esp32:
+  variant: esp32s3
+  framework:
+    type: esp-idf
+
+psram:
+  mode: octal
+
+binary_sensor:
+  - platform: gpio
+    pin: GPIO34
+    name: test

tests/component_tests/esp32/config/psram_quad_gpio34.yaml

@@ -0,0 +1,15 @@
+esphome:
+  name: test
+
+esp32:
+  variant: esp32s3
+  framework:
+    type: esp-idf
+
+psram:
+  mode: quad
+
+binary_sensor:
+  - platform: gpio
+    pin: GPIO34
+    name: test