diff --git a/.github/workflows/auto-label-pr.yml b/.github/workflows/auto-label-pr.yml
index 0e5ceb9346c..ea22f75ef04 100644
--- a/.github/workflows/auto-label-pr.yml
+++ b/.github/workflows/auto-label-pr.yml
@@ -27,9 +27,9 @@ jobs:
 
       - name: Generate a token
         id: generate-token
-        uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v2
+        uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
 
       - name: Auto Label PR
diff --git a/.github/workflows/lock.yml b/.github/workflows/lock.yml
index 20f9a74ea97..8d1dfe857de 100644
--- a/.github/workflows/lock.yml
+++ b/.github/workflows/lock.yml
@@ -8,4 +8,4 @@ on:
 
 jobs:
   lock:
-    uses: esphome/workflows/.github/workflows/lock.yml@3c4e8446aa1029f1c346a482034b3ee1489077ca  # 2026.4.0
+    uses: esphome/workflows/.github/workflows/lock.yml@025a1e6255610c498ed590403b7e510b69e474df  # 2026.4.1
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 35b9e065e1e..a16af92b6fb 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -223,7 +223,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: home-assistant-addon
@@ -258,7 +258,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: esphome-schema
@@ -289,7 +289,7 @@ jobs:
         id: generate-token
         uses: actions/create-github-app-token@1b10c78c7865c340bc4f6099eb2f838309f1e8c3 # v3.1.1
         with:
-          app-id: ${{ secrets.ESPHOME_GITHUB_APP_ID }}
+          client-id: ${{ vars.ESPHOME_GITHUB_APP_CLIENT_ID }}
           private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}
           owner: esphome
           repositories: version-notifier
diff --git a/esphome/components/esp32/__init__.py b/esphome/components/esp32/__init__.py
index eb023ce32c3..bd299d71f05 100644
--- a/esphome/components/esp32/__init__.py
+++ b/esphome/components/esp32/__init__.py
@@ -729,6 +729,9 @@ ESP_IDF_FRAMEWORK_VERSION_LOOKUP = {
     "dev": cv.Version(5, 5, 4),
 }
 ESP_IDF_PLATFORM_VERSION_LOOKUP = {
+    cv.Version(
+        6, 0, 1
+    ): "https://github.com/pioarduino/platform-espressif32.git#prep_IDF6",
     cv.Version(
         6, 0, 0
     ): "https://github.com/pioarduino/platform-espressif32.git#prep_IDF6",
diff --git a/esphome/components/esp8266/core.cpp b/esphome/components/esp8266/core.cpp
index 9161ca6aaf5..b9ad4082e98 100644
--- a/esphome/components/esp8266/core.cpp
+++ b/esphome/components/esp8266/core.cpp
@@ -3,98 +3,12 @@
 #include "core.h"
 #include "esphome/core/defines.h"
 #include "esphome/core/hal.h"
-#include "esphome/core/time_64.h"
-#include "esphome/core/helpers.h"
-#include "preferences.h"
 #include <Arduino.h>
-#include <core_esp8266_features.h>
-
-extern "C" {
-#include <user_interface.h>
-}
 
 namespace esphome {
 
-// yield(), micros(), millis_64() inlined in hal.h.
-// Fast accumulator replacement for Arduino's millis() (~3.3 μs via 4× 64-bit
-// multiplies on the LX106). Tracks a running ms counter from 32-bit
-// system_get_time() deltas using pure 32-bit ops. Installed as __wrap_millis
-// (via -Wl,--wrap=millis) so Arduino libs and IRAM_ATTR ISR handlers (e.g.
-// Wiegand, ZyAura) also get the fast version. xt_rsil(15) guards the static
-// state against ISR re-entry; the critical section is bounded (≤10 while-loop
-// iterations, ~100 ns on the common path, or a constant-time /1000 ~2.5 μs on
-// the rare path — well under WiFi's ~10 μs ISR latency budget). NMIs (level
-// >15) are not masked, but the ESP8266 SDK's NMI handlers don't call millis().
-//
-// system_get_time() wraps every ~71.6 min; unsigned (now_us - last_us) handles
-// one wrap. The main loop calls millis() at 60+ Hz, so delta stays tiny — a
-// >71 min block would trip the watchdog long before it could matter here.
-static constexpr uint32_t MILLIS_RARE_PATH_THRESHOLD_US = 10000;
-static constexpr uint32_t US_PER_MS = 1000;
-
-uint32_t IRAM_ATTR HOT millis() {
-  // Struct packs the three statics so the compiler loads one base address
-  // instead of three separate literal pool entries (saves ~8 bytes IRAM).
-  static struct {
-    uint32_t cache;
-    uint32_t remainder;
-    uint32_t last_us;
-  } state = {0, 0, 0};
-  uint32_t ps = xt_rsil(15);
-  uint32_t now_us = system_get_time();
-  uint32_t delta = now_us - state.last_us;
-  state.last_us = now_us;
-  state.remainder += delta;
-  if (state.remainder >= MILLIS_RARE_PATH_THRESHOLD_US) {
-    // Rare path: large gap (WiFi scan, boot, long block). Constant-time
-    // conversion keeps the critical section bounded.
-    uint32_t ms = state.remainder / US_PER_MS;
-    state.cache += ms;
-    // Reuse ms instead of `remainder %= US_PER_MS` — `%` would compile to a
-    // second __umodsi3 call on the LX106 (no hardware divide).
-    state.remainder -= ms * US_PER_MS;
-  } else {
-    // Common path: small gap. At most ~10 iterations since remainder was
-    // < threshold (10 ms) on entry and delta adds at most one more threshold
-    // before exiting this branch.
-    while (state.remainder >= US_PER_MS) {
-      state.cache++;
-      state.remainder -= US_PER_MS;
-    }
-  }
-  uint32_t result = state.cache;
-  xt_wsr_ps(ps);
-  return result;
-}
-// Poll-based delay that avoids ::delay() — Arduino's __delay has an intra-object
-// call to the original millis() that --wrap can't intercept, so calling ::delay()
-// would keep the slow Arduino millis body alive in IRAM. optimistic_yield still
-// enters esp_schedule()/esp_suspend_within_cont() via yield(), so SDK tasks and
-// WiFi run correctly. Theoretically less power-efficient than Arduino's
-// os_timer-based delay() for long waits, but nearly all ESPHome delays are short
-// (sensor/I²C/SPI settling in the 1–100 ms range) where the difference is
-// negligible.
-void HOT delay(uint32_t ms) {
-  if (ms == 0) {
-    optimistic_yield(1000);
-    return;
-  }
-  uint32_t start = millis();
-  while (millis() - start < ms) {
-    optimistic_yield(1000);
-  }
-}
-// delayMicroseconds(), arch_feed_wdt(), and progmem_read_*() are inlined in hal/hal_esp8266.h.
-void arch_restart() {
-  system_restart();
-  // restart() doesn't always end execution
-  while (true) {  // NOLINT(clang-diagnostic-unreachable-code)
-    yield();
-  }
-}
-void arch_init() {}
-uint32_t IRAM_ATTR HOT arch_get_cpu_cycle_count() { return esp_get_cycle_count(); }
-uint32_t arch_get_cpu_freq_hz() { return F_CPU; }
+// HAL functions live in hal.cpp. This file keeps only the ESP8266-specific
+// firmware bootstrap (Tasmota OTA magic bytes, optional GPIO pre-init).
 
 void force_link_symbols() {
   // Tasmota uses magic bytes in the binary to check if an OTA firmware is compatible
@@ -131,12 +45,4 @@ extern "C" void resetPins() {  // NOLINT
 
 }  // namespace esphome
 
-// Linker wrap: redirect all ::millis() calls (Arduino libs, ISRs) to our accumulator.
-// Requires -Wl,--wrap=millis in build flags (added by __init__.py).
-// NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
-extern "C" uint32_t IRAM_ATTR __wrap_millis() { return esphome::millis(); }
-// Note: Arduino's init() registers a 60-second overflow timer for micros64().
-// We leave it running — wrapping init() as a no-op would break micros64()'s
-// overflow tracking, and the timer's cost is negligible (~3 μs per 60 s).
-
 #endif  // USE_ESP8266
diff --git a/esphome/components/esp8266/hal.cpp b/esphome/components/esp8266/hal.cpp
new file mode 100644
index 00000000000..56910e5b399
--- /dev/null
+++ b/esphome/components/esp8266/hal.cpp
@@ -0,0 +1,111 @@
+#ifdef USE_ESP8266
+
+#include "esphome/core/hal.h"
+#include "esphome/core/helpers.h"
+
+#include <Arduino.h>
+#include <core_esp8266_features.h>
+
+extern "C" {
+#include <user_interface.h>
+}
+
+// Empty esp8266 namespace block to satisfy ci-custom's lint_namespace check.
+// HAL functions live in namespace esphome (root) — they are not part of the
+// esp8266 component's API.
+namespace esphome::esp8266 {}  // namespace esphome::esp8266
+
+namespace esphome {
+
+// yield(), micros(), millis_64(), delayMicroseconds(), arch_feed_wdt(),
+// progmem_read_*() are inlined in core/hal/hal_esp8266.h.
+//
+// Fast accumulator replacement for Arduino's millis() (~3.3 μs via 4× 64-bit
+// multiplies on the LX106). Tracks a running ms counter from 32-bit
+// system_get_time() deltas using pure 32-bit ops. Installed as __wrap_millis
+// (via -Wl,--wrap=millis) so Arduino libs and IRAM_ATTR ISR handlers (e.g.
+// Wiegand, ZyAura) also get the fast version. xt_rsil(15) guards the static
+// state against ISR re-entry; the critical section is bounded (≤10 while-loop
+// iterations, ~100 ns on the common path, or a constant-time /1000 ~2.5 μs on
+// the rare path — well under WiFi's ~10 μs ISR latency budget). NMIs (level
+// >15) are not masked, but the ESP8266 SDK's NMI handlers don't call millis().
+//
+// system_get_time() wraps every ~71.6 min; unsigned (now_us - last_us) handles
+// one wrap. The main loop calls millis() at 60+ Hz, so delta stays tiny — a
+// >71 min block would trip the watchdog long before it could matter here.
+static constexpr uint32_t MILLIS_RARE_PATH_THRESHOLD_US = 10000;
+static constexpr uint32_t US_PER_MS = 1000;
+
+uint32_t IRAM_ATTR HOT millis() {
+  // Struct packs the three statics so the compiler loads one base address
+  // instead of three separate literal pool entries (saves ~8 bytes IRAM).
+  static struct {
+    uint32_t cache;
+    uint32_t remainder;
+    uint32_t last_us;
+  } state = {0, 0, 0};
+  uint32_t ps = xt_rsil(15);
+  uint32_t now_us = system_get_time();
+  uint32_t delta = now_us - state.last_us;
+  state.last_us = now_us;
+  state.remainder += delta;
+  if (state.remainder >= MILLIS_RARE_PATH_THRESHOLD_US) {
+    // Rare path: large gap (WiFi scan, boot, long block). Constant-time
+    // conversion keeps the critical section bounded.
+    uint32_t ms = state.remainder / US_PER_MS;
+    state.cache += ms;
+    // Reuse ms instead of `remainder %= US_PER_MS` — `%` would compile to a
+    // second __umodsi3 call on the LX106 (no hardware divide).
+    state.remainder -= ms * US_PER_MS;
+  } else {
+    // Common path: small gap. At most ~10 iterations since remainder was
+    // < threshold (10 ms) on entry and delta adds at most one more threshold
+    // before exiting this branch.
+    while (state.remainder >= US_PER_MS) {
+      state.cache++;
+      state.remainder -= US_PER_MS;
+    }
+  }
+  uint32_t result = state.cache;
+  xt_wsr_ps(ps);
+  return result;
+}
+
+// Poll-based delay that avoids ::delay() — Arduino's __delay has an intra-object
+// call to the original millis() that --wrap can't intercept, so calling ::delay()
+// would keep the slow Arduino millis body alive in IRAM. optimistic_yield still
+// enters esp_schedule()/esp_suspend_within_cont() via yield(), so SDK tasks and
+// WiFi run correctly. Theoretically less power-efficient than Arduino's
+// os_timer-based delay() for long waits, but nearly all ESPHome delays are short
+// (sensor/I²C/SPI settling in the 1–100 ms range) where the difference is
+// negligible.
+void HOT delay(uint32_t ms) {
+  if (ms == 0) {
+    optimistic_yield(1000);
+    return;
+  }
+  uint32_t start = millis();
+  while (millis() - start < ms) {
+    optimistic_yield(1000);
+  }
+}
+
+void arch_restart() {
+  system_restart();
+  // restart() doesn't always end execution
+  while (true) {  // NOLINT(clang-diagnostic-unreachable-code)
+    yield();
+  }
+}
+
+}  // namespace esphome
+
+// Linker wrap: redirect all ::millis() calls (Arduino libs, ISRs) to our accumulator.
+// Requires -Wl,--wrap=millis in build flags (added by __init__.py).
+// NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
+extern "C" uint32_t IRAM_ATTR __wrap_millis() { return esphome::millis(); }
+// Note: Arduino's init() registers a 60-second overflow timer for micros64().
+// We leave it running — wrapping init() as a no-op would break micros64()'s
+// overflow tracking, and the timer's cost is negligible (~3 μs per 60 s).
+
+#endif  // USE_ESP8266
diff --git a/esphome/components/mcp23xxx_base/__init__.py b/esphome/components/mcp23xxx_base/__init__.py
index cd952099c05..76a3aabe3fa 100644
--- a/esphome/components/mcp23xxx_base/__init__.py
+++ b/esphome/components/mcp23xxx_base/__init__.py
@@ -2,6 +2,7 @@ from esphome import pins
 import esphome.codegen as cg
 import esphome.config_validation as cv
 from esphome.const import (
+    CONF_ALLOW_OTHER_USES,
     CONF_ID,
     CONF_INPUT,
     CONF_INTERRUPT,
@@ -30,10 +31,29 @@ MCP23XXX_INTERRUPT_MODES = {
     "FALLING": MCP23XXXInterruptMode.MCP23XXX_FALLING,
 }
 
+
+def _validate_interrupt_pin(value):
+    # The MCP component owns INT polarity (active-low, hardcoded falling-edge ISR)
+    # and installs a single ISR per GPIO, so neither inversion nor sharing is supported.
+    value = pins.internal_gpio_input_pin_schema(value)
+    if value.get(CONF_INVERTED):
+        raise cv.Invalid(
+            f"'{CONF_INVERTED}: true' is not supported on '{CONF_INTERRUPT_PIN}'; "
+            "the MCP23xxx INT line is fixed active-low"
+        )
+    if value.get(CONF_ALLOW_OTHER_USES):
+        raise cv.Invalid(
+            f"'{CONF_ALLOW_OTHER_USES}: true' is not supported on '{CONF_INTERRUPT_PIN}'; "
+            "sharing the interrupt pin between multiple MCP23xxx (or other components) "
+            "is not implemented. Remove the interrupt_pin to fall back to polling."
+        )
+    return value
+
+
 MCP23XXX_CONFIG_SCHEMA = cv.Schema(
     {
         cv.Optional(CONF_OPEN_DRAIN_INTERRUPT, default=False): cv.boolean,
-        cv.Optional(CONF_INTERRUPT_PIN): pins.internal_gpio_input_pin_schema,
+        cv.Optional(CONF_INTERRUPT_PIN): _validate_interrupt_pin,
     }
 ).extend(cv.COMPONENT_SCHEMA)
 
diff --git a/esphome/components/st7789v/__init__.py b/esphome/components/st7789v/__init__.py
index 3e64d09c57f..7915cf119c9 100644
--- a/esphome/components/st7789v/__init__.py
+++ b/esphome/components/st7789v/__init__.py
@@ -1,3 +1,8 @@
 import esphome.codegen as cg
 
 st7789v_ns = cg.esphome_ns.namespace("st7789v")
+
+DEPRECATED_COMPONENT = """
+The 'st7789v' component is deprecated and no new functionality will be added to it.
+PRs should target the newer and more performant 'mipi_spi' component.
+"""
diff --git a/esphome/components/st7789v/display.py b/esphome/components/st7789v/display.py
index 745c37f47dd..3b4d6d99ea5 100644
--- a/esphome/components/st7789v/display.py
+++ b/esphome/components/st7789v/display.py
@@ -1,3 +1,5 @@
+import logging
+
 from esphome import pins
 import esphome.codegen as cg
 from esphome.components import display, power_supply, spi
@@ -26,6 +28,8 @@ CODEOWNERS = ["@kbx81"]
 
 DEPENDENCIES = ["spi"]
 
+LOGGER = logging.getLogger(__name__)
+
 ST7789V = st7789v_ns.class_(
     "ST7789V", cg.PollingComponent, spi.SPIDevice, display.DisplayBuffer
 )
@@ -175,6 +179,9 @@ FINAL_VALIDATE_SCHEMA = spi.final_validate_device_schema(
 
 
 async def to_code(config):
+    LOGGER.warning(
+        "The 'st7789v' component is deprecated, it is recommended to use 'mipi_spi' instead."
+    )
     var = cg.new_Pvariable(config[CONF_ID])
     await display.register_display(var, config)
     await spi.register_spi_device(var, config, write_only=True)
diff --git a/esphome/components/sx126x/__init__.py b/esphome/components/sx126x/__init__.py
index b8696158fe5..a4ba5c34f31 100644
--- a/esphome/components/sx126x/__init__.py
+++ b/esphome/components/sx126x/__init__.py
@@ -1,3 +1,5 @@
+from typing import Any
+
 from esphome import automation, pins
 import esphome.codegen as cg
 from esphome.components import spi
@@ -5,6 +7,8 @@ from esphome.components.const import CONF_CRC_ENABLE, CONF_ON_PACKET
 import esphome.config_validation as cv
 from esphome.const import CONF_BUSY_PIN, CONF_DATA, CONF_FREQUENCY, CONF_ID
 from esphome.core import ID, TimePeriod
+from esphome.cpp_generator import MockObj
+from esphome.types import ConfigType, TemplateArgsType
 
 MULTI_CONF = True
 CODEOWNERS = ["@swoboda1337"]
@@ -15,6 +19,7 @@ CONF_SX126X_ID = "sx126x_id"
 CONF_BANDWIDTH = "bandwidth"
 CONF_BITRATE = "bitrate"
 CONF_CODING_RATE = "coding_rate"
+CONF_COLD = "cold"
 CONF_CRC_INVERTED = "crc_inverted"
 CONF_CRC_SIZE = "crc_size"
 CONF_CRC_POLYNOMIAL = "crc_polynomial"
@@ -144,7 +149,7 @@ SetModeStandbyAction = sx126x_ns.class_(
 )
 
 
-def validate_raw_data(value):
+def validate_raw_data(value: Any) -> bytes | list[int]:
     if isinstance(value, str):
         return value.encode("utf-8")
     if isinstance(value, list):
@@ -154,7 +159,7 @@ def validate_raw_data(value):
     )
 
 
-def validate_config(config):
+def validate_config(config: ConfigType) -> ConfigType:
     lora_bws = [
         "7_8kHz",
         "10_4kHz",
@@ -235,7 +240,7 @@ CONFIG_SCHEMA = (
 )
 
 
-async def to_code(config):
+async def to_code(config: ConfigType) -> None:
     var = cg.new_Pvariable(config[CONF_ID])
     await cg.register_component(var, config)
     await spi.register_spi_device(var, config)
@@ -307,24 +312,50 @@ NO_ARGS_ACTION_SCHEMA = automation.maybe_simple_id(
     NO_ARGS_ACTION_SCHEMA,
     synchronous=True,
 )
-@automation.register_action(
-    "sx126x.set_mode_sleep",
-    SetModeSleepAction,
-    NO_ARGS_ACTION_SCHEMA,
-    synchronous=True,
-)
 @automation.register_action(
     "sx126x.set_mode_standby",
     SetModeStandbyAction,
     NO_ARGS_ACTION_SCHEMA,
     synchronous=True,
 )
-async def no_args_action_to_code(config, action_id, template_arg, args):
+async def no_args_action_to_code(
+    config: ConfigType,
+    action_id: ID,
+    template_arg: cg.TemplateArguments,
+    args: TemplateArgsType,
+) -> MockObj:
     var = cg.new_Pvariable(action_id, template_arg)
     await cg.register_parented(var, config[CONF_ID])
     return var
 
 
+SET_MODE_SLEEP_ACTION_SCHEMA = automation.maybe_simple_id(
+    {
+        cv.GenerateID(): cv.use_id(SX126x),
+        cv.Optional(CONF_COLD, default=False): cv.templatable(cv.boolean),
+    }
+)
+
+
+@automation.register_action(
+    "sx126x.set_mode_sleep",
+    SetModeSleepAction,
+    SET_MODE_SLEEP_ACTION_SCHEMA,
+    synchronous=True,
+)
+async def set_mode_sleep_action_to_code(
+    config: ConfigType,
+    action_id: ID,
+    template_arg: cg.TemplateArguments,
+    args: TemplateArgsType,
+) -> MockObj:
+    var = cg.new_Pvariable(action_id, template_arg)
+    await cg.register_parented(var, config[CONF_ID])
+    template_ = await cg.templatable(config[CONF_COLD], args, bool)
+    cg.add(var.set_cold(template_))
+    return var
+
+
 SEND_PACKET_ACTION_SCHEMA = cv.maybe_simple_value(
     {
         cv.GenerateID(): cv.use_id(SX126x),
@@ -340,7 +371,12 @@ SEND_PACKET_ACTION_SCHEMA = cv.maybe_simple_value(
     SEND_PACKET_ACTION_SCHEMA,
     synchronous=True,
 )
-async def send_packet_action_to_code(config, action_id, template_arg, args):
+async def send_packet_action_to_code(
+    config: ConfigType,
+    action_id: ID,
+    template_arg: cg.TemplateArguments,
+    args: TemplateArgsType,
+) -> MockObj:
     var = cg.new_Pvariable(action_id, template_arg)
     await cg.register_parented(var, config[CONF_ID])
     data = config[CONF_DATA]
diff --git a/esphome/components/sx126x/automation.h b/esphome/components/sx126x/automation.h
index 2282c583cbf..ed5986e0974 100644
--- a/esphome/components/sx126x/automation.h
+++ b/esphome/components/sx126x/automation.h
@@ -56,7 +56,8 @@ template<typename... Ts> class SetModeRxAction : public Action<Ts...>, public Pa
 
 template<typename... Ts> class SetModeSleepAction : public Action<Ts...>, public Parented<SX126x> {
  public:
-  void play(const Ts &...x) override { this->parent_->set_mode_sleep(); }
+  TEMPLATABLE_VALUE(bool, cold)
+  void play(const Ts &...x) override { this->parent_->set_mode_sleep(this->cold_.value(x...)); }
 };
 
 template<typename... Ts> class SetModeStandbyAction : public Action<Ts...>, public Parented<SX126x> {
diff --git a/esphome/components/sx126x/sx126x.cpp b/esphome/components/sx126x/sx126x.cpp
index 6ea09e3a9e8..02f7d972a99 100644
--- a/esphome/components/sx126x/sx126x.cpp
+++ b/esphome/components/sx126x/sx126x.cpp
@@ -459,9 +459,10 @@ void SX126x::set_mode_tx() {
   this->write_opcode_(RADIO_SET_TX, buf, 3);
 }
 
-void SX126x::set_mode_sleep() {
+void SX126x::set_mode_sleep(bool cold) {
+  // 0x04 = warm start (config retained), 0x00 = cold start (config lost, lowest power)
   uint8_t buf[1];
-  buf[0] = 0x05;
+  buf[0] = cold ? 0x00 : 0x04;
   this->write_opcode_(RADIO_SET_SLEEP, buf, 1);
 }
 
diff --git a/esphome/components/sx126x/sx126x.h b/esphome/components/sx126x/sx126x.h
index edc00e3727e..87bbf18c793 100644
--- a/esphome/components/sx126x/sx126x.h
+++ b/esphome/components/sx126x/sx126x.h
@@ -79,7 +79,7 @@ class SX126x : public Component,
   void set_mode_rx();
   void set_mode_tx();
   void set_mode_standby(SX126xStandbyMode mode);
-  void set_mode_sleep();
+  void set_mode_sleep(bool cold = false);
   void set_modulation(uint8_t modulation) { this->modulation_ = modulation; }
   void set_pa_power(int8_t power) { this->pa_power_ = power; }
   void set_pa_ramp(uint8_t ramp) { this->pa_ramp_ = ramp; }
diff --git a/esphome/core/component.h b/esphome/core/component.h
index 6afcfda41db..185d51ab37d 100644
--- a/esphome/core/component.h
+++ b/esphome/core/component.h
@@ -655,7 +655,15 @@ class WarnIfComponentBlockingGuard {
   // Inlined: the fast path is just millis() + subtract + compare
   inline uint32_t HOT finish() {
 #ifdef USE_RUNTIME_STATS
-    this->component_->runtime_stats_.record_time(micros() - this->started_us_);
+    uint32_t elapsed_us = micros() - this->started_us_;
+    // component_ is nullptr for self-keyed scheduler items (set_timeout/set_interval(self, ...))
+    if (this->component_ != nullptr) {
+      this->component_->runtime_stats_.record_time(elapsed_us);
+    } else {
+      // Still accumulate into the global counter so Application::loop() can subtract
+      // this time from before_loop_tasks_ wall time.
+      ComponentRuntimeStats::global_recorded_us += elapsed_us;
+    }
 #endif
     uint32_t curr_time = MillisInternal::get();
 #ifndef USE_BENCHMARK
diff --git a/esphome/core/hal.h b/esphome/core/hal.h
index 312effa7b0e..a53296979c6 100644
--- a/esphome/core/hal.h
+++ b/esphome/core/hal.h
@@ -31,11 +31,10 @@
 namespace esphome {
 
 // Cross-platform declarations. delayMicroseconds(), arch_feed_wdt(),
-// arch_get_cpu_cycle_count() vary per platform (some inline, some
-// out-of-line) so they live in hal/hal_<platform>.h.
+// arch_get_cpu_cycle_count(), arch_init(), arch_get_cpu_freq_hz() vary
+// per platform (some inline, some out-of-line) so they live in
+// hal/hal_<platform>.h.
 void __attribute__((noreturn)) arch_restart();
-void arch_init();
-uint32_t arch_get_cpu_freq_hz();
 
 #ifndef USE_ESP8266
 // All non-ESP8266 platforms: PROGMEM is a no-op, so these are direct dereferences.
diff --git a/esphome/core/hal/hal_esp32.h b/esphome/core/hal/hal_esp32.h
index 2bff4244419..d5d7752bf6a 100644
--- a/esphome/core/hal/hal_esp32.h
+++ b/esphome/core/hal/hal_esp32.h
@@ -42,6 +42,9 @@ __attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { dela
 __attribute__((always_inline)) inline void arch_feed_wdt() { esp_task_wdt_reset(); }
 __attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return esp_cpu_get_cycle_count(); }
 
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
+
 }  // namespace esphome
 
 #endif  // USE_ESP32
diff --git a/esphome/core/hal/hal_esp8266.h b/esphome/core/hal/hal_esp8266.h
index 484118f1f50..b6e3b1ee3cb 100644
--- a/esphome/core/hal/hal_esp8266.h
+++ b/esphome/core/hal/hal_esp8266.h
@@ -3,6 +3,7 @@
 #ifdef USE_ESP8266
 
 #include <c_types.h>
+#include <core_esp8266_features.h>
 #include <cstdint>
 #include <pgmspace.h>
 
@@ -59,8 +60,11 @@ __attribute__((always_inline)) inline uint16_t progmem_read_uint16(const uint16_
 // NOLINTNEXTLINE(readability-identifier-naming)
 __attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
 __attribute__((always_inline)) inline void arch_feed_wdt() { system_soft_wdt_feed(); }
-
-uint32_t arch_get_cpu_cycle_count();
+__attribute__((always_inline)) inline void arch_init() {}
+// esp_get_cycle_count() declared in <core_esp8266_features.h>; F_CPU is a
+// compiler-driven macro from the ESP8266 Arduino board defs (-DF_CPU=...).
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_cycle_count() { return esp_get_cycle_count(); }
+__attribute__((always_inline)) inline uint32_t arch_get_cpu_freq_hz() { return F_CPU; }
 
 }  // namespace esphome
 
diff --git a/esphome/core/hal/hal_host.h b/esphome/core/hal/hal_host.h
index 682a1a422b4..a8896fdf638 100644
--- a/esphome/core/hal/hal_host.h
+++ b/esphome/core/hal/hal_host.h
@@ -22,6 +22,8 @@ uint64_t millis_64();
 void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
 void arch_feed_wdt();
 uint32_t arch_get_cpu_cycle_count();
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
 
 }  // namespace esphome
 
diff --git a/esphome/core/hal/hal_libretiny.h b/esphome/core/hal/hal_libretiny.h
index e9d33b7753b..ecfe830fe35 100644
--- a/esphome/core/hal/hal_libretiny.h
+++ b/esphome/core/hal/hal_libretiny.h
@@ -91,6 +91,8 @@ __attribute__((always_inline)) inline uint64_t millis_64() { return Millis64Impl
 void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
 void arch_feed_wdt();
 uint32_t arch_get_cpu_cycle_count();
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
 
 }  // namespace esphome
 
diff --git a/esphome/core/hal/hal_rp2040.h b/esphome/core/hal/hal_rp2040.h
index 2a1d67b4a32..46f6e421cd4 100644
--- a/esphome/core/hal/hal_rp2040.h
+++ b/esphome/core/hal/hal_rp2040.h
@@ -38,6 +38,8 @@ __attribute__((always_inline)) inline uint64_t millis_64() { return micros_to_mi
 void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
 void arch_feed_wdt();
 uint32_t arch_get_cpu_cycle_count();
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
 
 }  // namespace esphome
 
diff --git a/esphome/core/hal/hal_zephyr.h b/esphome/core/hal/hal_zephyr.h
index 6707c85b2c2..d4b37b5eb6c 100644
--- a/esphome/core/hal/hal_zephyr.h
+++ b/esphome/core/hal/hal_zephyr.h
@@ -22,6 +22,8 @@ uint64_t millis_64();
 void delayMicroseconds(uint32_t us);  // NOLINT(readability-identifier-naming)
 void arch_feed_wdt();
 uint32_t arch_get_cpu_cycle_count();
+void arch_init();
+uint32_t arch_get_cpu_freq_hz();
 
 }  // namespace esphome
 
diff --git a/esphome/platformio_api.py b/esphome/platformio_api.py
index dec541985f6..c0cd048890a 100644
--- a/esphome/platformio_api.py
+++ b/esphome/platformio_api.py
@@ -14,6 +14,37 @@ from esphome.util import run_external_process
 _LOGGER = logging.getLogger(__name__)
 
 
+def _strip_win_long_path_prefix(path: str) -> str:
+    r"""Strip the Windows extended-length path prefix from ``path``.
+
+    Handles both forms documented at
+    https://learn.microsoft.com/windows/win32/fileio/naming-a-file:
+
+    * ``\\?\C:\path\to\file`` -> ``C:\path\to\file``
+    * ``\\?\UNC\server\share\path`` -> ``\\server\share\path``
+
+    The NSIS-installed ``esphome.exe`` launcher on Windows starts Python with
+    ``sys.executable`` already prefixed with ``\\?\``. That prefix propagates
+    into PlatformIO's ``$PYTHONEXE`` (PlatformIO reads ``PYTHONEXEPATH`` from
+    the environment, falling back to ``os.path.normpath(sys.executable)``)
+    and ends up baked into SCons-emitted command lines for build steps such
+    as the esp8266 ``elf2bin`` invocation. ``cmd.exe`` does not understand
+    the ``\\?\`` prefix, so the build fails with
+    "The system cannot find the path specified." Stripping the prefix early
+    keeps the path shell-quotable.
+
+    No-op on non-Windows platforms.
+    """
+    if sys.platform != "win32":
+        return path
+    if path.startswith("\\\\?\\UNC\\"):
+        # \\?\UNC\server\share\... -> \\server\share\...
+        return "\\\\" + path[len("\\\\?\\UNC\\") :]
+    if path.startswith("\\\\?\\"):
+        return path[len("\\\\?\\") :]
+    return path
+
+
 def run_platformio_cli(*args, **kwargs) -> str | int:
     os.environ["PLATFORMIO_FORCE_COLOR"] = "true"
     os.environ["PLATFORMIO_BUILD_DIR"] = str(CORE.relative_pioenvs_path().absolute())
@@ -24,7 +55,18 @@ def run_platformio_cli(*args, **kwargs) -> str | int:
     os.environ.setdefault("PYTHONWARNINGS", "ignore::SyntaxWarning")
     # Increase uv retry count to handle transient network errors (default is 3)
     os.environ.setdefault("UV_HTTP_RETRIES", "10")
-    cmd = [sys.executable, "-m", "esphome.platformio_runner"] + list(args)
+    # Strip the Windows extended-length path prefix from sys.executable so it
+    # doesn't propagate into PlatformIO's $PYTHONEXE and break SCons-emitted
+    # command lines run through cmd.exe.
+    python_exe = _strip_win_long_path_prefix(sys.executable)
+    if python_exe != sys.executable:
+        # Only override PYTHONEXEPATH when we actually stripped a prefix.
+        # PlatformIO's get_pythonexe_path() reads this and falls back to
+        # sys.executable otherwise; setting it unconditionally would clobber
+        # a user-provided value (or the unmodified path on platforms that
+        # don't need the strip).
+        os.environ["PYTHONEXEPATH"] = python_exe
+    cmd = [python_exe, "-m", "esphome.platformio_runner"] + list(args)
 
     return run_external_process(*cmd, **kwargs)
 
diff --git a/requirements.txt b/requirements.txt
index abc8ac5dbb6..789a3f79957 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -12,7 +12,7 @@ platformio==6.1.19
 esptool==5.2.0
 click==8.3.3
 esphome-dashboard==20260425.0
-aioesphomeapi==44.22.0
+aioesphomeapi==44.23.0
 zeroconf==0.148.0
 puremagic==1.30
 ruamel.yaml==0.19.1 # dashboard_import
diff --git a/tests/unit_tests/test_platformio_api.py b/tests/unit_tests/test_platformio_api.py
index 67e64e5f612..b241622f89d 100644
--- a/tests/unit_tests/test_platformio_api.py
+++ b/tests/unit_tests/test_platformio_api.py
@@ -311,6 +311,105 @@ def test_run_platformio_cli_sets_environment_variables(
         assert "arg" in args
 
 
+@pytest.mark.parametrize(
+    ("platform", "input_path", "expected"),
+    [
+        # win32: drive-letter extended-length prefix is stripped
+        (
+            "win32",
+            "\\\\?\\C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+        ),
+        # win32: UNC extended-length prefix is translated to a regular UNC path
+        (
+            "win32",
+            "\\\\?\\UNC\\server\\share\\python.exe",
+            "\\\\server\\share\\python.exe",
+        ),
+        # win32: paths without the prefix are returned unchanged
+        (
+            "win32",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+            "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe",
+        ),
+        # non-win32: prefix is left alone (no-op)
+        ("linux", "\\\\?\\C:\\python.exe", "\\\\?\\C:\\python.exe"),
+        ("darwin", "/usr/bin/python3", "/usr/bin/python3"),
+    ],
+)
+def test_strip_win_long_path_prefix(
+    platform: str, input_path: str, expected: str
+) -> None:
+    r"""``\\?\`` and ``\\?\UNC\`` prefixes are stripped only on win32."""
+    with patch("esphome.platformio_api.sys.platform", platform):
+        assert platformio_api._strip_win_long_path_prefix(input_path) == expected
+
+
+def test_run_platformio_cli_strips_win_long_path_prefix(
+    setup_core: Path, mock_run_external_process: Mock
+) -> None:
+    r"""Windows ``\\?\`` prefix on sys.executable does not leak into the subprocess.
+
+    The NSIS-installed esphome.exe launcher starts Python with
+    ``sys.executable`` already prefixed by the extended-length path marker.
+    That prefix would otherwise propagate into PlatformIO's ``PYTHONEXE`` and
+    break SCons-emitted command lines run through ``cmd.exe``.
+    """
+    CORE.build_path = str(setup_core / "build" / "test")
+    prefixed_exe = (
+        "\\\\?\\C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe"
+    )
+    stripped_exe = (
+        "C:\\Users\\jesse\\AppData\\Local\\ESPHome Builder\\python\\python.exe"
+    )
+
+    with (
+        patch.dict(os.environ, {}, clear=False),
+        patch("esphome.platformio_api.sys.platform", "win32"),
+        patch("esphome.platformio_api.sys.executable", prefixed_exe),
+    ):
+        # Pop any pre-existing PYTHONEXEPATH so the assertion below reflects
+        # what run_platformio_cli set, not whatever the test runner's
+        # environment happened to contain.
+        os.environ.pop("PYTHONEXEPATH", None)
+        mock_run_external_process.return_value = 0
+        platformio_api.run_platformio_cli("test", "arg")
+
+        # The subprocess is invoked with the stripped executable path.
+        mock_run_external_process.assert_called_once()
+        args = mock_run_external_process.call_args[0]
+        assert args[0] == stripped_exe
+        # PYTHONEXEPATH is exported with the stripped path so PlatformIO's
+        # get_pythonexe_path() picks it up in the subprocess.
+        assert os.environ["PYTHONEXEPATH"] == stripped_exe
+
+
+def test_run_platformio_cli_does_not_set_pythonexepath_without_strip(
+    setup_core: Path, mock_run_external_process: Mock
+) -> None:
+    r"""PYTHONEXEPATH is not touched when sys.executable has no ``\\?\`` prefix.
+
+    Setting it unconditionally would clobber a user-provided value (or
+    interfere with non-Windows tooling that has no prefix to strip).
+    """
+    CORE.build_path = str(setup_core / "build" / "test")
+    plain_exe = "/usr/bin/python3"
+
+    with (
+        patch.dict(os.environ, {}, clear=False),
+        patch("esphome.platformio_api.sys.platform", "linux"),
+        patch("esphome.platformio_api.sys.executable", plain_exe),
+    ):
+        os.environ.pop("PYTHONEXEPATH", None)
+        mock_run_external_process.return_value = 0
+        platformio_api.run_platformio_cli("test", "arg")
+
+        mock_run_external_process.assert_called_once()
+        args = mock_run_external_process.call_args[0]
+        assert args[0] == plain_exe
+        assert "PYTHONEXEPATH" not in os.environ
+
+
 def test_run_platformio_cli_run_builds_command(
     setup_core: Path, mock_run_platformio_cli: Mock
 ) -> None: