diff --git a/esphome/components/api/api_connection.h b/esphome/components/api/api_connection.h
index 3873c7fcac..9ed18c24dc 100644
--- a/esphome/components/api/api_connection.h
+++ b/esphome/components/api/api_connection.h
@@ -301,8 +301,10 @@ class APIConnection : public APIServerConnection {
     if (entity->has_own_name())
       msg.name = entity->get_name();
 
-    // Set common EntityBase properties
+      // Set common EntityBase properties
+#ifdef USE_ENTITY_ICON
     msg.icon = entity->get_icon();
+#endif
     msg.disabled_by_default = entity->is_disabled_by_default();
     msg.entity_category = static_cast<enums::EntityCategory>(entity->get_entity_category());
 #ifdef USE_DEVICES
diff --git a/esphome/components/api/api_pb2.h b/esphome/components/api/api_pb2.h
index 39f00b4adc..95db58aae9 100644
--- a/esphome/components/api/api_pb2.h
+++ b/esphome/components/api/api_pb2.h
@@ -292,9 +292,13 @@ class InfoResponseProtoMessage : public ProtoMessage {
   uint32_t key{0};
   std::string name{};
   bool disabled_by_default{false};
+#ifdef USE_ENTITY_ICON
   std::string icon{};
+#endif
   enums::EntityCategory entity_category{};
+#ifdef USE_DEVICES
   uint32_t device_id{0};
+#endif
 
  protected:
 };
@@ -303,7 +307,9 @@ class StateResponseProtoMessage : public ProtoMessage {
  public:
   ~StateResponseProtoMessage() override = default;
   uint32_t key{0};
+#ifdef USE_DEVICES
   uint32_t device_id{0};
+#endif
 
  protected:
 };
@@ -312,7 +318,9 @@ class CommandProtoMessage : public ProtoDecodableMessage {
  public:
   ~CommandProtoMessage() override = default;
   uint32_t key{0};
+#ifdef USE_DEVICES
   uint32_t device_id{0};
+#endif
 
  protected:
 };
diff --git a/esphome/components/bluetooth_proxy/bluetooth_connection.cpp b/esphome/components/bluetooth_proxy/bluetooth_connection.cpp
index 2bfccdb438..dae6e521bb 100644
--- a/esphome/components/bluetooth_proxy/bluetooth_connection.cpp
+++ b/esphome/components/bluetooth_proxy/bluetooth_connection.cpp
@@ -13,11 +13,180 @@ namespace bluetooth_proxy {
 
 static const char *const TAG = "bluetooth_proxy.connection";
 
+static std::vector<uint64_t> get_128bit_uuid_vec(esp_bt_uuid_t uuid_source) {
+  esp_bt_uuid_t uuid = espbt::ESPBTUUID::from_uuid(uuid_source).as_128bit().get_uuid();
+  return std::vector<uint64_t>{((uint64_t) uuid.uuid.uuid128[15] << 56) | ((uint64_t) uuid.uuid.uuid128[14] << 48) |
+                                   ((uint64_t) uuid.uuid.uuid128[13] << 40) | ((uint64_t) uuid.uuid.uuid128[12] << 32) |
+                                   ((uint64_t) uuid.uuid.uuid128[11] << 24) | ((uint64_t) uuid.uuid.uuid128[10] << 16) |
+                                   ((uint64_t) uuid.uuid.uuid128[9] << 8) | ((uint64_t) uuid.uuid.uuid128[8]),
+                               ((uint64_t) uuid.uuid.uuid128[7] << 56) | ((uint64_t) uuid.uuid.uuid128[6] << 48) |
+                                   ((uint64_t) uuid.uuid.uuid128[5] << 40) | ((uint64_t) uuid.uuid.uuid128[4] << 32) |
+                                   ((uint64_t) uuid.uuid.uuid128[3] << 24) | ((uint64_t) uuid.uuid.uuid128[2] << 16) |
+                                   ((uint64_t) uuid.uuid.uuid128[1] << 8) | ((uint64_t) uuid.uuid.uuid128[0])};
+}
+
 void BluetoothConnection::dump_config() {
   ESP_LOGCONFIG(TAG, "BLE Connection:");
   BLEClientBase::dump_config();
 }
 
+void BluetoothConnection::loop() {
+  BLEClientBase::loop();
+
+  // Early return if no active connection or not in service discovery phase
+  if (this->address_ == 0 || this->send_service_ < 0 || this->send_service_ > this->service_count_) {
+    return;
+  }
+
+  // Handle service discovery
+  this->send_service_for_discovery_();
+}
+
+void BluetoothConnection::reset_connection_(esp_err_t reason) {
+  // Send disconnection notification
+  this->proxy_->send_device_connection(this->address_, false, 0, reason);
+
+  // Important: If we were in the middle of sending services, we do NOT send
+  // send_gatt_services_done() here. This ensures the client knows that
+  // the service discovery was interrupted and can retry. The client
+  // (aioesphomeapi) implements a 30-second timeout (DEFAULT_BLE_TIMEOUT)
+  // to detect incomplete service discovery rather than relying on us to
+  // tell them about a partial list.
+  this->set_address(0);
+  this->send_service_ = DONE_SENDING_SERVICES;
+  this->proxy_->send_connections_free();
+}
+
+void BluetoothConnection::send_service_for_discovery_() {
+  if (this->send_service_ == this->service_count_) {
+    this->send_service_ = DONE_SENDING_SERVICES;
+    this->proxy_->send_gatt_services_done(this->address_);
+    if (this->connection_type_ == espbt::ConnectionType::V3_WITH_CACHE ||
+        this->connection_type_ == espbt::ConnectionType::V3_WITHOUT_CACHE) {
+      this->release_services();
+    }
+    return;
+  }
+
+  // Early return if no API connection
+  auto *api_conn = this->proxy_->get_api_connection();
+  if (api_conn == nullptr) {
+    return;
+  }
+
+  // Send next service
+  esp_gattc_service_elem_t service_result;
+  uint16_t service_count = 1;
+  esp_gatt_status_t service_status = esp_ble_gattc_get_service(this->gattc_if_, this->conn_id_, nullptr,
+                                                               &service_result, &service_count, this->send_service_);
+  this->send_service_++;
+
+  if (service_status != ESP_GATT_OK) {
+    ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_service error at offset=%d, status=%d", this->connection_index_,
+             this->address_str().c_str(), this->send_service_ - 1, service_status);
+    return;
+  }
+
+  if (service_count == 0) {
+    ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_service missing, service_count=%d", this->connection_index_,
+             this->address_str().c_str(), service_count);
+    return;
+  }
+
+  api::BluetoothGATTGetServicesResponse resp;
+  resp.address = this->address_;
+  resp.services.reserve(1);  // Always one service per response in this implementation
+  api::BluetoothGATTService service_resp;
+  service_resp.uuid = get_128bit_uuid_vec(service_result.uuid);
+  service_resp.handle = service_result.start_handle;
+
+  // Get the number of characteristics directly with one call
+  uint16_t total_char_count = 0;
+  esp_gatt_status_t char_count_status =
+      esp_ble_gattc_get_attr_count(this->gattc_if_, this->conn_id_, ESP_GATT_DB_CHARACTERISTIC,
+                                   service_result.start_handle, service_result.end_handle, 0, &total_char_count);
+
+  if (char_count_status == ESP_GATT_OK && total_char_count > 0) {
+    // Only reserve if we successfully got a count
+    service_resp.characteristics.reserve(total_char_count);
+  } else if (char_count_status != ESP_GATT_OK) {
+    ESP_LOGW(TAG, "[%d] [%s] Error getting characteristic count, status=%d", this->connection_index_,
+             this->address_str().c_str(), char_count_status);
+  }
+
+  // Now process characteristics
+  uint16_t char_offset = 0;
+  esp_gattc_char_elem_t char_result;
+  while (true) {  // characteristics
+    uint16_t char_count = 1;
+    esp_gatt_status_t char_status =
+        esp_ble_gattc_get_all_char(this->gattc_if_, this->conn_id_, service_result.start_handle,
+                                   service_result.end_handle, &char_result, &char_count, char_offset);
+    if (char_status == ESP_GATT_INVALID_OFFSET || char_status == ESP_GATT_NOT_FOUND) {
+      break;
+    }
+    if (char_status != ESP_GATT_OK) {
+      ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_all_char error, status=%d", this->connection_index_,
+               this->address_str().c_str(), char_status);
+      break;
+    }
+    if (char_count == 0) {
+      break;
+    }
+
+    api::BluetoothGATTCharacteristic characteristic_resp;
+    characteristic_resp.uuid = get_128bit_uuid_vec(char_result.uuid);
+    characteristic_resp.handle = char_result.char_handle;
+    characteristic_resp.properties = char_result.properties;
+    char_offset++;
+
+    // Get the number of descriptors directly with one call
+    uint16_t total_desc_count = 0;
+    esp_gatt_status_t desc_count_status =
+        esp_ble_gattc_get_attr_count(this->gattc_if_, this->conn_id_, ESP_GATT_DB_DESCRIPTOR, char_result.char_handle,
+                                     service_result.end_handle, 0, &total_desc_count);
+
+    if (desc_count_status == ESP_GATT_OK && total_desc_count > 0) {
+      // Only reserve if we successfully got a count
+      characteristic_resp.descriptors.reserve(total_desc_count);
+    } else if (desc_count_status != ESP_GATT_OK) {
+      ESP_LOGW(TAG, "[%d] [%s] Error getting descriptor count for char handle %d, status=%d", this->connection_index_,
+               this->address_str().c_str(), char_result.char_handle, desc_count_status);
+    }
+
+    // Now process descriptors
+    uint16_t desc_offset = 0;
+    esp_gattc_descr_elem_t desc_result;
+    while (true) {  // descriptors
+      uint16_t desc_count = 1;
+      esp_gatt_status_t desc_status = esp_ble_gattc_get_all_descr(
+          this->gattc_if_, this->conn_id_, char_result.char_handle, &desc_result, &desc_count, desc_offset);
+      if (desc_status == ESP_GATT_INVALID_OFFSET || desc_status == ESP_GATT_NOT_FOUND) {
+        break;
+      }
+      if (desc_status != ESP_GATT_OK) {
+        ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_all_descr error, status=%d", this->connection_index_,
+                 this->address_str().c_str(), desc_status);
+        break;
+      }
+      if (desc_count == 0) {
+        break;
+      }
+
+      api::BluetoothGATTDescriptor descriptor_resp;
+      descriptor_resp.uuid = get_128bit_uuid_vec(desc_result.uuid);
+      descriptor_resp.handle = desc_result.handle;
+      characteristic_resp.descriptors.push_back(std::move(descriptor_resp));
+      desc_offset++;
+    }
+    service_resp.characteristics.push_back(std::move(characteristic_resp));
+  }
+  resp.services.push_back(std::move(service_resp));
+
+  // Send the message (we already checked api_conn is not null at the beginning)
+  api_conn->send_message(resp, api::BluetoothGATTGetServicesResponse::MESSAGE_TYPE);
+}
+
 bool BluetoothConnection::gattc_event_handler(esp_gattc_cb_event_t event, esp_gatt_if_t gattc_if,
                                               esp_ble_gattc_cb_param_t *param) {
   if (!BLEClientBase::gattc_event_handler(event, gattc_if, param))
@@ -25,22 +194,16 @@ bool BluetoothConnection::gattc_event_handler(esp_gattc_cb_event_t event, esp_ga
 
   switch (event) {
     case ESP_GATTC_DISCONNECT_EVT: {
-      this->proxy_->send_device_connection(this->address_, false, 0, param->disconnect.reason);
-      this->set_address(0);
-      this->proxy_->send_connections_free();
+      this->reset_connection_(param->disconnect.reason);
       break;
     }
     case ESP_GATTC_CLOSE_EVT: {
-      this->proxy_->send_device_connection(this->address_, false, 0, param->close.reason);
-      this->set_address(0);
-      this->proxy_->send_connections_free();
+      this->reset_connection_(param->close.reason);
       break;
     }
     case ESP_GATTC_OPEN_EVT: {
       if (param->open.status != ESP_GATT_OK && param->open.status != ESP_GATT_ALREADY_OPEN) {
-        this->proxy_->send_device_connection(this->address_, false, 0, param->open.status);
-        this->set_address(0);
-        this->proxy_->send_connections_free();
+        this->reset_connection_(param->open.status);
       } else if (this->connection_type_ == espbt::ConnectionType::V3_WITH_CACHE) {
         this->proxy_->send_device_connection(this->address_, true, this->mtu_);
         this->proxy_->send_connections_free();
diff --git a/esphome/components/bluetooth_proxy/bluetooth_connection.h b/esphome/components/bluetooth_proxy/bluetooth_connection.h
index 73c034d93b..2673238fba 100644
--- a/esphome/components/bluetooth_proxy/bluetooth_connection.h
+++ b/esphome/components/bluetooth_proxy/bluetooth_connection.h
@@ -12,6 +12,7 @@ class BluetoothProxy;
 class BluetoothConnection : public esp32_ble_client::BLEClientBase {
  public:
   void dump_config() override;
+  void loop() override;
   bool gattc_event_handler(esp_gattc_cb_event_t event, esp_gatt_if_t gattc_if,
                            esp_ble_gattc_cb_param_t *param) override;
   void gap_event_handler(esp_gap_ble_cb_event_t event, esp_ble_gap_cb_param_t *param) override;
@@ -27,6 +28,9 @@ class BluetoothConnection : public esp32_ble_client::BLEClientBase {
  protected:
   friend class BluetoothProxy;
 
+  void send_service_for_discovery_();
+  void reset_connection_(esp_err_t reason);
+
   // Memory optimized layout for 32-bit systems
   // Group 1: Pointers (4 bytes each, naturally aligned)
   BluetoothProxy *proxy_;
diff --git a/esphome/components/bluetooth_proxy/bluetooth_proxy.cpp b/esphome/components/bluetooth_proxy/bluetooth_proxy.cpp
index 7d12842a24..f4b63f3a5d 100644
--- a/esphome/components/bluetooth_proxy/bluetooth_proxy.cpp
+++ b/esphome/components/bluetooth_proxy/bluetooth_proxy.cpp
@@ -11,19 +11,6 @@ namespace esphome {
 namespace bluetooth_proxy {
 
 static const char *const TAG = "bluetooth_proxy";
-static const int DONE_SENDING_SERVICES = -2;
-
-std::vector<uint64_t> get_128bit_uuid_vec(esp_bt_uuid_t uuid_source) {
-  esp_bt_uuid_t uuid = espbt::ESPBTUUID::from_uuid(uuid_source).as_128bit().get_uuid();
-  return std::vector<uint64_t>{((uint64_t) uuid.uuid.uuid128[15] << 56) | ((uint64_t) uuid.uuid.uuid128[14] << 48) |
-                                   ((uint64_t) uuid.uuid.uuid128[13] << 40) | ((uint64_t) uuid.uuid.uuid128[12] << 32) |
-                                   ((uint64_t) uuid.uuid.uuid128[11] << 24) | ((uint64_t) uuid.uuid.uuid128[10] << 16) |
-                                   ((uint64_t) uuid.uuid.uuid128[9] << 8) | ((uint64_t) uuid.uuid.uuid128[8]),
-                               ((uint64_t) uuid.uuid.uuid128[7] << 56) | ((uint64_t) uuid.uuid.uuid128[6] << 48) |
-                                   ((uint64_t) uuid.uuid.uuid128[5] << 40) | ((uint64_t) uuid.uuid.uuid128[4] << 32) |
-                                   ((uint64_t) uuid.uuid.uuid128[3] << 24) | ((uint64_t) uuid.uuid.uuid128[2] << 16) |
-                                   ((uint64_t) uuid.uuid.uuid128[1] << 8) | ((uint64_t) uuid.uuid.uuid128[0])};
-}
 
 // Batch size for BLE advertisements to maximize WiFi efficiency
 // Each advertisement is up to 80 bytes when packaged (including protocol overhead)
@@ -213,130 +200,12 @@ void BluetoothProxy::loop() {
   }
 
   // Flush any pending BLE advertisements that have been accumulated but not yet sent
-  static uint32_t last_flush_time = 0;
   uint32_t now = App.get_loop_component_start_time();
 
   // Flush accumulated advertisements every 100ms
-  if (now - last_flush_time >= 100) {
+  if (now - this->last_advertisement_flush_time_ >= 100) {
     this->flush_pending_advertisements();
-    last_flush_time = now;
-  }
-  for (auto *connection : this->connections_) {
-    if (connection->send_service_ == connection->service_count_) {
-      connection->send_service_ = DONE_SENDING_SERVICES;
-      this->send_gatt_services_done(connection->get_address());
-      if (connection->connection_type_ == espbt::ConnectionType::V3_WITH_CACHE ||
-          connection->connection_type_ == espbt::ConnectionType::V3_WITHOUT_CACHE) {
-        connection->release_services();
-      }
-    } else if (connection->send_service_ >= 0) {
-      esp_gattc_service_elem_t service_result;
-      uint16_t service_count = 1;
-      esp_gatt_status_t service_status =
-          esp_ble_gattc_get_service(connection->get_gattc_if(), connection->get_conn_id(), nullptr, &service_result,
-                                    &service_count, connection->send_service_);
-      connection->send_service_++;
-      if (service_status != ESP_GATT_OK) {
-        ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_service error at offset=%d, status=%d",
-                 connection->get_connection_index(), connection->address_str().c_str(), connection->send_service_ - 1,
-                 service_status);
-        continue;
-      }
-      if (service_count == 0) {
-        ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_service missing, service_count=%d",
-                 connection->get_connection_index(), connection->address_str().c_str(), service_count);
-        continue;
-      }
-      api::BluetoothGATTGetServicesResponse resp;
-      resp.address = connection->get_address();
-      resp.services.reserve(1);  // Always one service per response in this implementation
-      api::BluetoothGATTService service_resp;
-      service_resp.uuid = get_128bit_uuid_vec(service_result.uuid);
-      service_resp.handle = service_result.start_handle;
-      uint16_t char_offset = 0;
-      esp_gattc_char_elem_t char_result;
-      // Get the number of characteristics directly with one call
-      uint16_t total_char_count = 0;
-      esp_gatt_status_t char_count_status = esp_ble_gattc_get_attr_count(
-          connection->get_gattc_if(), connection->get_conn_id(), ESP_GATT_DB_CHARACTERISTIC,
-          service_result.start_handle, service_result.end_handle, 0, &total_char_count);
-
-      if (char_count_status == ESP_GATT_OK && total_char_count > 0) {
-        // Only reserve if we successfully got a count
-        service_resp.characteristics.reserve(total_char_count);
-      } else if (char_count_status != ESP_GATT_OK) {
-        ESP_LOGW(TAG, "[%d] [%s] Error getting characteristic count, status=%d", connection->get_connection_index(),
-                 connection->address_str().c_str(), char_count_status);
-      }
-
-      // Now process characteristics
-      while (true) {  // characteristics
-        uint16_t char_count = 1;
-        esp_gatt_status_t char_status = esp_ble_gattc_get_all_char(
-            connection->get_gattc_if(), connection->get_conn_id(), service_result.start_handle,
-            service_result.end_handle, &char_result, &char_count, char_offset);
-        if (char_status == ESP_GATT_INVALID_OFFSET || char_status == ESP_GATT_NOT_FOUND) {
-          break;
-        }
-        if (char_status != ESP_GATT_OK) {
-          ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_all_char error, status=%d", connection->get_connection_index(),
-                   connection->address_str().c_str(), char_status);
-          break;
-        }
-        if (char_count == 0) {
-          break;
-        }
-        api::BluetoothGATTCharacteristic characteristic_resp;
-        characteristic_resp.uuid = get_128bit_uuid_vec(char_result.uuid);
-        characteristic_resp.handle = char_result.char_handle;
-        characteristic_resp.properties = char_result.properties;
-        char_offset++;
-
-        // Get the number of descriptors directly with one call
-        uint16_t total_desc_count = 0;
-        esp_gatt_status_t desc_count_status =
-            esp_ble_gattc_get_attr_count(connection->get_gattc_if(), connection->get_conn_id(), ESP_GATT_DB_DESCRIPTOR,
-                                         char_result.char_handle, service_result.end_handle, 0, &total_desc_count);
-
-        if (desc_count_status == ESP_GATT_OK && total_desc_count > 0) {
-          // Only reserve if we successfully got a count
-          characteristic_resp.descriptors.reserve(total_desc_count);
-        } else if (desc_count_status != ESP_GATT_OK) {
-          ESP_LOGW(TAG, "[%d] [%s] Error getting descriptor count for char handle %d, status=%d",
-                   connection->get_connection_index(), connection->address_str().c_str(), char_result.char_handle,
-                   desc_count_status);
-        }
-
-        // Now process descriptors
-        uint16_t desc_offset = 0;
-        esp_gattc_descr_elem_t desc_result;
-        while (true) {  // descriptors
-          uint16_t desc_count = 1;
-          esp_gatt_status_t desc_status =
-              esp_ble_gattc_get_all_descr(connection->get_gattc_if(), connection->get_conn_id(),
-                                          char_result.char_handle, &desc_result, &desc_count, desc_offset);
-          if (desc_status == ESP_GATT_INVALID_OFFSET || desc_status == ESP_GATT_NOT_FOUND) {
-            break;
-          }
-          if (desc_status != ESP_GATT_OK) {
-            ESP_LOGE(TAG, "[%d] [%s] esp_ble_gattc_get_all_descr error, status=%d", connection->get_connection_index(),
-                     connection->address_str().c_str(), desc_status);
-            break;
-          }
-          if (desc_count == 0) {
-            break;
-          }
-          api::BluetoothGATTDescriptor descriptor_resp;
-          descriptor_resp.uuid = get_128bit_uuid_vec(desc_result.uuid);
-          descriptor_resp.handle = desc_result.handle;
-          characteristic_resp.descriptors.push_back(std::move(descriptor_resp));
-          desc_offset++;
-        }
-        service_resp.characteristics.push_back(std::move(characteristic_resp));
-      }
-      resp.services.push_back(std::move(service_resp));
-      this->api_connection_->send_message(resp, api::BluetoothGATTGetServicesResponse::MESSAGE_TYPE);
-    }
+    this->last_advertisement_flush_time_ = now;
   }
 }
 
diff --git a/esphome/components/bluetooth_proxy/bluetooth_proxy.h b/esphome/components/bluetooth_proxy/bluetooth_proxy.h
index 52f1d0f88a..9d84a9dbf2 100644
--- a/esphome/components/bluetooth_proxy/bluetooth_proxy.h
+++ b/esphome/components/bluetooth_proxy/bluetooth_proxy.h
@@ -22,6 +22,7 @@ namespace esphome {
 namespace bluetooth_proxy {
 
 static const esp_err_t ESP_GATT_NOT_CONNECTED = -1;
+static const int DONE_SENDING_SERVICES = -2;
 
 using namespace esp32_ble_client;
 
@@ -149,7 +150,10 @@ class BluetoothProxy : public esp32_ble_tracker::ESPBTDeviceListener, public Com
   std::vector<api::BluetoothLERawAdvertisement> advertisement_pool_;
   std::unique_ptr<api::BluetoothLERawAdvertisementsResponse> response_;
 
-  // Group 3: 1-byte types grouped together
+  // Group 3: 4-byte types
+  uint32_t last_advertisement_flush_time_{0};
+
+  // Group 4: 1-byte types grouped together
   bool active_;
   uint8_t advertisement_count_{0};
   // 2 bytes used, 2 bytes padding
diff --git a/esphome/components/esp32/__init__.py b/esphome/components/esp32/__init__.py
index c772a3438c..6ddb579733 100644
--- a/esphome/components/esp32/__init__.py
+++ b/esphome/components/esp32/__init__.py
@@ -31,6 +31,7 @@ from esphome.const import (
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
     PLATFORM_ESP32,
+    CoreModel,
     __version__,
 )
 from esphome.core import CORE, HexInt, TimePeriod
@@ -713,6 +714,7 @@ async def to_code(config):
     cg.add_define("ESPHOME_BOARD", config[CONF_BOARD])
     cg.add_build_flag(f"-DUSE_ESP32_VARIANT_{config[CONF_VARIANT]}")
     cg.add_define("ESPHOME_VARIANT", VARIANT_FRIENDLY[config[CONF_VARIANT]])
+    cg.add_define(CoreModel.MULTI_ATOMICS)
 
     cg.add_platformio_option("lib_ldf_mode", "off")
     cg.add_platformio_option("lib_compat_mode", "strict")
diff --git a/esphome/components/esp32_ble_tracker/esp32_ble_tracker.cpp b/esphome/components/esp32_ble_tracker/esp32_ble_tracker.cpp
index 44577afbbd..96003073d7 100644
--- a/esphome/components/esp32_ble_tracker/esp32_ble_tracker.cpp
+++ b/esphome/components/esp32_ble_tracker/esp32_ble_tracker.cpp
@@ -128,46 +128,53 @@ void ESP32BLETracker::loop() {
     uint8_t write_idx = this->ring_write_index_.load(std::memory_order_acquire);
 
     while (read_idx != write_idx) {
-      // Process one result at a time directly from ring buffer
-      BLEScanResult &scan_result = this->scan_ring_buffer_[read_idx];
+      // Calculate how many contiguous results we can process in one batch
+      // If write > read: process all results from read to write
+      // If write <= read (wraparound): process from read to end of buffer first
+      size_t batch_size = (write_idx > read_idx) ? (write_idx - read_idx) : (SCAN_RESULT_BUFFER_SIZE - read_idx);
 
+      // Process the batch for raw advertisements
       if (this->raw_advertisements_) {
         for (auto *listener : this->listeners_) {
-          listener->parse_devices(&scan_result, 1);
+          listener->parse_devices(&this->scan_ring_buffer_[read_idx], batch_size);
         }
         for (auto *client : this->clients_) {
-          client->parse_devices(&scan_result, 1);
+          client->parse_devices(&this->scan_ring_buffer_[read_idx], batch_size);
         }
       }
 
+      // Process individual results for parsed advertisements
       if (this->parse_advertisements_) {
 #ifdef USE_ESP32_BLE_DEVICE
-        ESPBTDevice device;
-        device.parse_scan_rst(scan_result);
+        for (size_t i = 0; i < batch_size; i++) {
+          BLEScanResult &scan_result = this->scan_ring_buffer_[read_idx + i];
+          ESPBTDevice device;
+          device.parse_scan_rst(scan_result);
 
-        bool found = false;
-        for (auto *listener : this->listeners_) {
-          if (listener->parse_device(device))
-            found = true;
-        }
+          bool found = false;
+          for (auto *listener : this->listeners_) {
+            if (listener->parse_device(device))
+              found = true;
+          }
 
-        for (auto *client : this->clients_) {
-          if (client->parse_device(device)) {
-            found = true;
-            if (!connecting && client->state() == ClientState::DISCOVERED) {
-              promote_to_connecting = true;
+          for (auto *client : this->clients_) {
+            if (client->parse_device(device)) {
+              found = true;
+              if (!connecting && client->state() == ClientState::DISCOVERED) {
+                promote_to_connecting = true;
+              }
             }
           }
-        }
 
-        if (!found && !this->scan_continuous_) {
-          this->print_bt_device_info(device);
+          if (!found && !this->scan_continuous_) {
+            this->print_bt_device_info(device);
+          }
         }
 #endif  // USE_ESP32_BLE_DEVICE
       }
 
-      // Move to next entry in ring buffer
-      read_idx = (read_idx + 1) % SCAN_RESULT_BUFFER_SIZE;
+      // Update read index for entire batch
+      read_idx = (read_idx + batch_size) % SCAN_RESULT_BUFFER_SIZE;
 
       // Store with release to ensure reads complete before index update
       this->ring_read_index_.store(read_idx, std::memory_order_release);
diff --git a/esphome/components/esp8266/__init__.py b/esphome/components/esp8266/__init__.py
index 01b20bdcb1..d08d7121b7 100644
--- a/esphome/components/esp8266/__init__.py
+++ b/esphome/components/esp8266/__init__.py
@@ -15,6 +15,7 @@ from esphome.const import (
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
     PLATFORM_ESP8266,
+    CoreModel,
 )
 from esphome.core import CORE, coroutine_with_priority
 from esphome.helpers import copy_file_if_changed
@@ -187,6 +188,7 @@ async def to_code(config):
     cg.set_cpp_standard("gnu++20")
     cg.add_define("ESPHOME_BOARD", config[CONF_BOARD])
     cg.add_define("ESPHOME_VARIANT", "ESP8266")
+    cg.add_define(CoreModel.SINGLE)
 
     cg.add_platformio_option("extra_scripts", ["post:post_build.py"])
 
diff --git a/esphome/components/host/__init__.py b/esphome/components/host/__init__.py
index a67d73fbb7..2d77f2f7ab 100644
--- a/esphome/components/host/__init__.py
+++ b/esphome/components/host/__init__.py
@@ -7,6 +7,7 @@ from esphome.const import (
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
     PLATFORM_HOST,
+    CoreModel,
 )
 from esphome.core import CORE
 
@@ -43,6 +44,7 @@ async def to_code(config):
     cg.add_define("USE_ESPHOME_HOST_MAC_ADDRESS", config[CONF_MAC_ADDRESS].parts)
     cg.add_build_flag("-std=gnu++20")
     cg.add_define("ESPHOME_BOARD", "host")
+    cg.add_define(CoreModel.MULTI_ATOMICS)
     cg.add_platformio_option("platform", "platformio/native")
     cg.add_platformio_option("lib_ldf_mode", "off")
     cg.add_platformio_option("lib_compat_mode", "strict")
diff --git a/esphome/components/libretiny/__init__.py b/esphome/components/libretiny/__init__.py
index 17d5d46ffd..7f2a0bc0a5 100644
--- a/esphome/components/libretiny/__init__.py
+++ b/esphome/components/libretiny/__init__.py
@@ -20,6 +20,7 @@ from esphome.const import (
     KEY_FRAMEWORK_VERSION,
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
+    CoreModel,
     __version__,
 )
 from esphome.core import CORE
@@ -260,6 +261,7 @@ async def component_to_code(config):
     cg.add_build_flag(f"-DUSE_LIBRETINY_VARIANT_{config[CONF_FAMILY]}")
     cg.add_define("ESPHOME_BOARD", config[CONF_BOARD])
     cg.add_define("ESPHOME_VARIANT", FAMILY_FRIENDLY[config[CONF_FAMILY]])
+    cg.add_define(CoreModel.MULTI_NO_ATOMICS)
 
     # force using arduino framework
     cg.add_platformio_option("framework", "arduino")
diff --git a/esphome/components/max7219digit/display.py b/esphome/components/max7219digit/display.py
index f195078c1a..fef121ff10 100644
--- a/esphome/components/max7219digit/display.py
+++ b/esphome/components/max7219digit/display.py
@@ -3,6 +3,7 @@ import esphome.codegen as cg
 from esphome.components import display, spi
 import esphome.config_validation as cv
 from esphome.const import (
+    CONF_FLIP_X,
     CONF_ID,
     CONF_INTENSITY,
     CONF_LAMBDA,
@@ -14,7 +15,6 @@ CODEOWNERS = ["@rspaargaren"]
 DEPENDENCIES = ["spi"]
 
 CONF_ROTATE_CHIP = "rotate_chip"
-CONF_FLIP_X = "flip_x"
 CONF_SCROLL_SPEED = "scroll_speed"
 CONF_SCROLL_DWELL = "scroll_dwell"
 CONF_SCROLL_DELAY = "scroll_delay"
diff --git a/esphome/components/rp2040/__init__.py b/esphome/components/rp2040/__init__.py
index 0fa299ce5c..28c3bbd70c 100644
--- a/esphome/components/rp2040/__init__.py
+++ b/esphome/components/rp2040/__init__.py
@@ -16,6 +16,7 @@ from esphome.const import (
     KEY_TARGET_FRAMEWORK,
     KEY_TARGET_PLATFORM,
     PLATFORM_RP2040,
+    CoreModel,
 )
 from esphome.core import CORE, EsphomeError, coroutine_with_priority
 from esphome.helpers import copy_file_if_changed, mkdir_p, read_file, write_file
@@ -171,6 +172,7 @@ async def to_code(config):
     cg.set_cpp_standard("gnu++20")
     cg.add_define("ESPHOME_BOARD", config[CONF_BOARD])
     cg.add_define("ESPHOME_VARIANT", "RP2040")
+    cg.add_define(CoreModel.SINGLE)
 
     cg.add_platformio_option("extra_scripts", ["post:post_build.py"])
 
diff --git a/esphome/components/ssd1306_base/__init__.py b/esphome/components/ssd1306_base/__init__.py
index ab2c7a5496..6633b24607 100644
--- a/esphome/components/ssd1306_base/__init__.py
+++ b/esphome/components/ssd1306_base/__init__.py
@@ -6,6 +6,8 @@ from esphome.const import (
     CONF_BRIGHTNESS,
     CONF_CONTRAST,
     CONF_EXTERNAL_VCC,
+    CONF_FLIP_X,
+    CONF_FLIP_Y,
     CONF_INVERT,
     CONF_LAMBDA,
     CONF_MODEL,
@@ -18,9 +20,6 @@ ssd1306_base_ns = cg.esphome_ns.namespace("ssd1306_base")
 SSD1306 = ssd1306_base_ns.class_("SSD1306", cg.PollingComponent, display.DisplayBuffer)
 SSD1306Model = ssd1306_base_ns.enum("SSD1306Model")
 
-CONF_FLIP_X = "flip_x"
-CONF_FLIP_Y = "flip_y"
-
 MODELS = {
     "SSD1306_128X32": SSD1306Model.SSD1306_MODEL_128_32,
     "SSD1306_128X64": SSD1306Model.SSD1306_MODEL_128_64,
diff --git a/esphome/const.py b/esphome/const.py
index 39578a1fcf..627b6bac18 100644
--- a/esphome/const.py
+++ b/esphome/const.py
@@ -35,6 +35,14 @@ class Framework(StrEnum):
     ZEPHYR = "zephyr"
 
 
+class CoreModel(StrEnum):
+    """Core model identifiers for ESPHome scheduler."""
+
+    SINGLE = "ESPHOME_CORES_SINGLE"
+    MULTI_NO_ATOMICS = "ESPHOME_CORES_MULTI_NO_ATOMICS"
+    MULTI_ATOMICS = "ESPHOME_CORES_MULTI_ATOMICS"
+
+
 class PlatformFramework(Enum):
     """Combined platform-framework identifiers with tuple values."""
 
@@ -375,6 +383,8 @@ CONF_FINGER_ID = "finger_id"
 CONF_FINGERPRINT_COUNT = "fingerprint_count"
 CONF_FLASH_LENGTH = "flash_length"
 CONF_FLASH_TRANSITION_LENGTH = "flash_transition_length"
+CONF_FLIP_X = "flip_x"
+CONF_FLIP_Y = "flip_y"
 CONF_FLOW = "flow"
 CONF_FLOW_CONTROL_PIN = "flow_control_pin"
 CONF_FONT = "font"
diff --git a/esphome/core/defines.h b/esphome/core/defines.h
index 7ddb3436cd..19d380bd29 100644
--- a/esphome/core/defines.h
+++ b/esphome/core/defines.h
@@ -15,6 +15,9 @@
 #define ESPHOME_VARIANT "ESP32"
 #define ESPHOME_DEBUG_SCHEDULER
 
+// Default threading model for static analysis (ESP32 is multi-core with atomics)
+#define ESPHOME_CORES_MULTI_ATOMICS
+
 // logger
 #define ESPHOME_LOG_LEVEL ESPHOME_LOG_LEVEL_VERY_VERBOSE
 
diff --git a/esphome/core/scheduler.cpp b/esphome/core/scheduler.cpp
index 7a0c08e1f0..d6d99f82c8 100644
--- a/esphome/core/scheduler.cpp
+++ b/esphome/core/scheduler.cpp
@@ -54,7 +54,7 @@ static void validate_static_string(const char *name) {
     ESP_LOGW(TAG, "WARNING: Scheduler name '%s' at %p might be on heap (static ref at %p)", name, name, static_str);
   }
 }
-#endif
+#endif /* ESPHOME_DEBUG_SCHEDULER */
 
 // A note on locking: the `lock_` lock protects the `items_` and `to_add_` containers. It must be taken when writing to
 // them (i.e. when adding/removing items, but not when changing items). As items are only deleted from the loop task,
@@ -82,9 +82,9 @@ void HOT Scheduler::set_timer_common_(Component *component, SchedulerItem::Type
   item->callback = std::move(func);
   item->remove = false;
 
-#if !defined(USE_ESP8266) && !defined(USE_RP2040)
+#ifndef ESPHOME_CORES_SINGLE
   // Special handling for defer() (delay = 0, type = TIMEOUT)
-  // ESP8266 and RP2040 are excluded because they don't need thread-safe defer handling
+  // Single-core platforms don't need thread-safe defer handling
   if (delay == 0 && type == SchedulerItem::TIMEOUT) {
     // Put in defer queue for guaranteed FIFO execution
     LockGuard guard{this->lock_};
@@ -92,7 +92,7 @@ void HOT Scheduler::set_timer_common_(Component *component, SchedulerItem::Type
     this->defer_queue_.push_back(std::move(item));
     return;
   }
-#endif
+#endif /* not ESPHOME_CORES_SINGLE */
 
   // Get fresh timestamp for new timer/interval - ensures accurate scheduling
   const auto now = this->millis_64_(millis());  // Fresh millis() call
@@ -123,7 +123,7 @@ void HOT Scheduler::set_timer_common_(Component *component, SchedulerItem::Type
     ESP_LOGD(TAG, "set_%s(name='%s/%s', %s=%" PRIu32 ", offset=%" PRIu32 ")", type_str, item->get_source(),
              name_cstr ? name_cstr : "(null)", type_str, delay, static_cast<uint32_t>(item->next_execution_ - now));
   }
-#endif
+#endif /* ESPHOME_DEBUG_SCHEDULER */
 
   LockGuard guard{this->lock_};
   // If name is provided, do atomic cancel-and-add
@@ -231,7 +231,7 @@ optional<uint32_t> HOT Scheduler::next_schedule_in(uint32_t now) {
   return item->next_execution_ - now_64;
 }
 void HOT Scheduler::call(uint32_t now) {
-#if !defined(USE_ESP8266) && !defined(USE_RP2040)
+#ifndef ESPHOME_CORES_SINGLE
   // Process defer queue first to guarantee FIFO execution order for deferred items.
   // Previously, defer() used the heap which gave undefined order for equal timestamps,
   // causing race conditions on multi-core systems (ESP32, BK7200).
@@ -239,8 +239,7 @@ void HOT Scheduler::call(uint32_t now) {
   // - Deferred items (delay=0) go directly to defer_queue_ in set_timer_common_
   // - Items execute in exact order they were deferred (FIFO guarantee)
   // - No deferred items exist in to_add_, so processing order doesn't affect correctness
-  // ESP8266 and RP2040 don't use this queue - they fall back to the heap-based approach
-  // (ESP8266: single-core, RP2040: empty mutex implementation).
+  // Single-core platforms don't use this queue and fall back to the heap-based approach.
   //
   // Note: Items cancelled via cancel_item_locked_() are marked with remove=true but still
   // processed here. They are removed from the queue normally via pop_front() but skipped
@@ -262,7 +261,7 @@ void HOT Scheduler::call(uint32_t now) {
       this->execute_item_(item.get(), now);
     }
   }
-#endif
+#endif /* not ESPHOME_CORES_SINGLE */
 
   // Convert the fresh timestamp from main loop to 64-bit for scheduler operations
   const auto now_64 = this->millis_64_(now);  // 'now' from parameter - fresh from Application::loop()
@@ -274,13 +273,15 @@ void HOT Scheduler::call(uint32_t now) {
   if (now_64 - last_print > 2000) {
     last_print = now_64;
     std::vector<std::unique_ptr<SchedulerItem>> old_items;
-#if !defined(USE_ESP8266) && !defined(USE_RP2040) && !defined(USE_LIBRETINY)
-    ESP_LOGD(TAG, "Items: count=%zu, now=%" PRIu64 " (%u, %" PRIu32 ")", this->items_.size(), now_64,
-             this->millis_major_, this->last_millis_.load(std::memory_order_relaxed));
-#else
-    ESP_LOGD(TAG, "Items: count=%zu, now=%" PRIu64 " (%u, %" PRIu32 ")", this->items_.size(), now_64,
+#ifdef ESPHOME_CORES_MULTI_ATOMICS
+    const auto last_dbg = this->last_millis_.load(std::memory_order_relaxed);
+    const auto major_dbg = this->millis_major_.load(std::memory_order_relaxed);
+    ESP_LOGD(TAG, "Items: count=%zu, now=%" PRIu64 " (%" PRIu16 ", %" PRIu32 ")", this->items_.size(), now_64,
+             major_dbg, last_dbg);
+#else  /* not ESPHOME_CORES_MULTI_ATOMICS */
+    ESP_LOGD(TAG, "Items: count=%zu, now=%" PRIu64 " (%" PRIu16 ", %" PRIu32 ")", this->items_.size(), now_64,
              this->millis_major_, this->last_millis_);
-#endif
+#endif /* else ESPHOME_CORES_MULTI_ATOMICS */
     while (!this->empty_()) {
       std::unique_ptr<SchedulerItem> item;
       {
@@ -305,7 +306,7 @@ void HOT Scheduler::call(uint32_t now) {
       std::make_heap(this->items_.begin(), this->items_.end(), SchedulerItem::cmp);
     }
   }
-#endif  // ESPHOME_DEBUG_SCHEDULER
+#endif /* ESPHOME_DEBUG_SCHEDULER */
 
   // If we have too many items to remove
   if (this->to_remove_ > MAX_LOGICALLY_DELETED_ITEMS) {
@@ -352,7 +353,7 @@ void HOT Scheduler::call(uint32_t now) {
       ESP_LOGV(TAG, "Running %s '%s/%s' with interval=%" PRIu32 " next_execution=%" PRIu64 " (now=%" PRIu64 ")",
                item->get_type_str(), item->get_source(), item_name ? item_name : "(null)", item->interval,
                item->next_execution_, now_64);
-#endif
+#endif /* ESPHOME_DEBUG_SCHEDULER */
 
       // Warning: During callback(), a lot of stuff can happen, including:
       //  - timeouts/intervals get added, potentially invalidating vector pointers
@@ -460,7 +461,7 @@ bool HOT Scheduler::cancel_item_locked_(Component *component, const char *name_c
   size_t total_cancelled = 0;
 
   // Check all containers for matching items
-#if !defined(USE_ESP8266) && !defined(USE_RP2040)
+#ifndef ESPHOME_CORES_SINGLE
   // Only check defer queue for timeouts (intervals never go there)
   if (type == SchedulerItem::TIMEOUT) {
     for (auto &item : this->defer_queue_) {
@@ -470,7 +471,7 @@ bool HOT Scheduler::cancel_item_locked_(Component *component, const char *name_c
       }
     }
   }
-#endif
+#endif /* not ESPHOME_CORES_SINGLE */
 
   // Cancel items in the main heap
   for (auto &item : this->items_) {
@@ -495,24 +496,53 @@ bool HOT Scheduler::cancel_item_locked_(Component *component, const char *name_c
 
 uint64_t Scheduler::millis_64_(uint32_t now) {
   // THREAD SAFETY NOTE:
-  // This function can be called from multiple threads simultaneously on ESP32/LibreTiny.
-  // On single-threaded platforms (ESP8266, RP2040), atomics are not needed.
+  // This function has three implementations, based on the precompiler flags
+  // - ESPHOME_CORES_SINGLE - Runs on single-core platforms (ESP8266, RP2040, etc.)
+  // - ESPHOME_CORES_MULTI_NO_ATOMICS - Runs on multi-core platforms without atomics (LibreTiny)
+  // - ESPHOME_CORES_MULTI_ATOMICS - Runs on multi-core platforms with atomics (ESP32, HOST, etc.)
+  //
+  // Make sure all changes are synchronized if you edit this function.
   //
   // IMPORTANT: Always pass fresh millis() values to this function. The implementation
   // handles out-of-order timestamps between threads, but minimizing time differences
   // helps maintain accuracy.
   //
-  // The implementation handles the 32-bit rollover (every 49.7 days) by:
-  // 1. Using a lock when detecting rollover to ensure atomic update
-  // 2. Restricting normal updates to forward movement within the same epoch
-  // This prevents race conditions at the rollover boundary without requiring
-  // 64-bit atomics or locking on every call.
 
-#ifdef USE_LIBRETINY
-  // LibreTiny: Multi-threaded but lacks atomic operation support
-  // TODO: If LibreTiny ever adds atomic support, remove this entire block and
-  // let it fall through to the atomic-based implementation below
-  // We need to use a lock when near the rollover boundary to prevent races
+#ifdef ESPHOME_CORES_SINGLE
+  // This is the single core implementation.
+  //
+  // Single-core platforms have no concurrency, so this is a simple implementation
+  // that just tracks 32-bit rollover (every 49.7 days) without any locking or atomics.
+
+  uint16_t major = this->millis_major_;
+  uint32_t last = this->last_millis_;
+
+  // Check for rollover
+  if (now < last && (last - now) > HALF_MAX_UINT32) {
+    this->millis_major_++;
+    major++;
+#ifdef ESPHOME_DEBUG_SCHEDULER
+    ESP_LOGD(TAG, "Detected true 32-bit rollover at %" PRIu32 "ms (was %" PRIu32 ")", now, last);
+#endif /* ESPHOME_DEBUG_SCHEDULER */
+  }
+
+  // Only update if time moved forward
+  if (now > last) {
+    this->last_millis_ = now;
+  }
+
+  // Combine major (high 32 bits) and now (low 32 bits) into 64-bit time
+  return now + (static_cast<uint64_t>(major) << 32);
+
+#elif defined(ESPHOME_CORES_MULTI_NO_ATOMICS)
+  // This is the multi core no atomics implementation.
+  //
+  // Without atomics, this implementation uses locks more aggressively:
+  // 1. Always locks when near the rollover boundary (within 10 seconds)
+  // 2. Always locks when detecting a large backwards jump
+  // 3. Updates without lock in normal forward progression (accepting minor races)
+  // This is less efficient but necessary without atomic operations.
+  uint16_t major = this->millis_major_;
   uint32_t last = this->last_millis_;
 
   // Define a safe window around the rollover point (10 seconds)
@@ -531,9 +561,10 @@ uint64_t Scheduler::millis_64_(uint32_t now) {
     if (now < last && (last - now) > HALF_MAX_UINT32) {
       // True rollover detected (happens every ~49.7 days)
       this->millis_major_++;
+      major++;
 #ifdef ESPHOME_DEBUG_SCHEDULER
       ESP_LOGD(TAG, "Detected true 32-bit rollover at %" PRIu32 "ms (was %" PRIu32 ")", now, last);
-#endif
+#endif /* ESPHOME_DEBUG_SCHEDULER */
     }
     // Update last_millis_ while holding lock
     this->last_millis_ = now;
@@ -549,58 +580,76 @@ uint64_t Scheduler::millis_64_(uint32_t now) {
   // If now <= last and we're not near rollover, don't update
   // This minimizes backwards time movement
 
-#elif !defined(USE_ESP8266) && !defined(USE_RP2040)
-  // Multi-threaded platforms with atomic support (ESP32)
-  uint32_t last = this->last_millis_.load(std::memory_order_relaxed);
+  // Combine major (high 32 bits) and now (low 32 bits) into 64-bit time
+  return now + (static_cast<uint64_t>(major) << 32);
 
-  // If we might be near a rollover (large backwards jump), take the lock for the entire operation
-  // This ensures rollover detection and last_millis_ update are atomic together
-  if (now < last && (last - now) > HALF_MAX_UINT32) {
-    // Potential rollover - need lock for atomic rollover detection + update
-    LockGuard guard{this->lock_};
-    // Re-read with lock held
-    last = this->last_millis_.load(std::memory_order_relaxed);
+#elif defined(ESPHOME_CORES_MULTI_ATOMICS)
+  // This is the multi core with atomics implementation.
+  //
+  // Uses atomic operations with acquire/release semantics to ensure coherent
+  // reads of millis_major_ and last_millis_ across cores. Features:
+  // 1. Epoch-coherency retry loop to handle concurrent updates
+  // 2. Lock only taken for actual rollover detection and update
+  // 3. Lock-free CAS updates for normal forward time progression
+  // 4. Memory ordering ensures cores see consistent time values
 
+  for (;;) {
+    uint16_t major = this->millis_major_.load(std::memory_order_acquire);
+
+    /*
+     * Acquire so that if we later decide **not** to take the lock we still
+     * observe a `millis_major_` value coherent with the loaded `last_millis_`.
+     * The acquire load ensures any later read of `millis_major_` sees its
+     * corresponding increment.
+     */
+    uint32_t last = this->last_millis_.load(std::memory_order_acquire);
+
+    // If we might be near a rollover (large backwards jump), take the lock for the entire operation
+    // This ensures rollover detection and last_millis_ update are atomic together
     if (now < last && (last - now) > HALF_MAX_UINT32) {
-      // True rollover detected (happens every ~49.7 days)
-      this->millis_major_++;
+      // Potential rollover - need lock for atomic rollover detection + update
+      LockGuard guard{this->lock_};
+      // Re-read with lock held; mutex already provides ordering
+      last = this->last_millis_.load(std::memory_order_relaxed);
+
+      if (now < last && (last - now) > HALF_MAX_UINT32) {
+        // True rollover detected (happens every ~49.7 days)
+        this->millis_major_.fetch_add(1, std::memory_order_relaxed);
+        major++;
 #ifdef ESPHOME_DEBUG_SCHEDULER
-      ESP_LOGD(TAG, "Detected true 32-bit rollover at %" PRIu32 "ms (was %" PRIu32 ")", now, last);
-#endif
-    }
-    // Update last_millis_ while holding lock to prevent races
-    this->last_millis_.store(now, std::memory_order_relaxed);
-  } else {
-    // Normal case: Try lock-free update, but only allow forward movement within same epoch
-    // This prevents accidentally moving backwards across a rollover boundary
-    while (now > last && (now - last) < HALF_MAX_UINT32) {
-      if (this->last_millis_.compare_exchange_weak(last, now, std::memory_order_relaxed)) {
-        break;
+        ESP_LOGD(TAG, "Detected true 32-bit rollover at %" PRIu32 "ms (was %" PRIu32 ")", now, last);
+#endif /* ESPHOME_DEBUG_SCHEDULER */
+      }
+      /*
+       * Update last_millis_ while holding the lock to prevent races
+       * Publish the new low-word *after* bumping `millis_major_` (done above)
+       * so readers never see a mismatched pair.
+       */
+      this->last_millis_.store(now, std::memory_order_release);
+    } else {
+      // Normal case: Try lock-free update, but only allow forward movement within same epoch
+      // This prevents accidentally moving backwards across a rollover boundary
+      while (now > last && (now - last) < HALF_MAX_UINT32) {
+        if (this->last_millis_.compare_exchange_weak(last, now,
+                                                     std::memory_order_release,     // success
+                                                     std::memory_order_relaxed)) {  // failure
+          break;
+        }
+        // CAS failure means no data was published; relaxed is fine
+        // last is automatically updated by compare_exchange_weak if it fails
       }
-      // last is automatically updated by compare_exchange_weak if it fails
     }
+    uint16_t major_end = this->millis_major_.load(std::memory_order_relaxed);
+    if (major_end == major)
+      return now + (static_cast<uint64_t>(major) << 32);
   }
+  // Unreachable - the loop always returns when major_end == major
+  __builtin_unreachable();
 
 #else
-  // Single-threaded platforms (ESP8266, RP2040): No atomics needed
-  uint32_t last = this->last_millis_;
-
-  // Check for rollover
-  if (now < last && (last - now) > HALF_MAX_UINT32) {
-    this->millis_major_++;
-#ifdef ESPHOME_DEBUG_SCHEDULER
-    ESP_LOGD(TAG, "Detected true 32-bit rollover at %" PRIu32 "ms (was %" PRIu32 ")", now, last);
+#error \
+    "No platform threading model defined. One of ESPHOME_CORES_SINGLE, ESPHOME_CORES_MULTI_NO_ATOMICS, or ESPHOME_CORES_MULTI_ATOMICS must be defined."
 #endif
-  }
-
-  // Only update if time moved forward
-  if (now > last) {
-    this->last_millis_ = now;
-  }
-#endif
-
-  // Combine major (high 32 bits) and now (low 32 bits) into 64-bit time
-  return now + (static_cast<uint64_t>(this->millis_major_) << 32);
 }
 
 bool HOT Scheduler::SchedulerItem::cmp(const std::unique_ptr<SchedulerItem> &a,
diff --git a/esphome/core/scheduler.h b/esphome/core/scheduler.h
index 64df2f2bb0..b539b26949 100644
--- a/esphome/core/scheduler.h
+++ b/esphome/core/scheduler.h
@@ -1,10 +1,11 @@
 #pragma once
 
+#include "esphome/core/defines.h"
 #include <vector>
 #include <memory>
 #include <cstring>
 #include <deque>
-#if !defined(USE_ESP8266) && !defined(USE_RP2040) && !defined(USE_LIBRETINY)
+#ifdef ESPHOME_CORES_MULTI_ATOMICS
 #include <atomic>
 #endif
 
@@ -204,23 +205,40 @@ class Scheduler {
   Mutex lock_;
   std::vector<std::unique_ptr<SchedulerItem>> items_;
   std::vector<std::unique_ptr<SchedulerItem>> to_add_;
-#if !defined(USE_ESP8266) && !defined(USE_RP2040)
-  // ESP8266 and RP2040 don't need the defer queue because:
-  // ESP8266: Single-core with no preemptive multitasking
-  // RP2040: Currently has empty mutex implementation in ESPHome
-  // Both platforms save 40 bytes of RAM by excluding this
+#ifndef ESPHOME_CORES_SINGLE
+  // Single-core platforms don't need the defer queue and save 40 bytes of RAM
   std::deque<std::unique_ptr<SchedulerItem>> defer_queue_;  // FIFO queue for defer() calls
-#endif
-#if !defined(USE_ESP8266) && !defined(USE_RP2040) && !defined(USE_LIBRETINY)
-  // Multi-threaded platforms with atomic support: last_millis_ needs atomic for lock-free updates
+#endif                                                      /* ESPHOME_CORES_SINGLE */
+  uint32_t to_remove_{0};
+
+#ifdef ESPHOME_CORES_MULTI_ATOMICS
+  /*
+   * Multi-threaded platforms with atomic support: last_millis_ needs atomic for lock-free updates
+   *
+   * MEMORY-ORDERING NOTE
+   * --------------------
+   * `last_millis_` and `millis_major_` form a single 64-bit timestamp split in half.
+   * Writers publish `last_millis_` with memory_order_release and readers use
+   * memory_order_acquire. This ensures that once a reader sees the new low word,
+   * it also observes the corresponding increment of `millis_major_`.
+   */
   std::atomic<uint32_t> last_millis_{0};
-#else
+#else  /* not ESPHOME_CORES_MULTI_ATOMICS */
   // Platforms without atomic support or single-threaded platforms
   uint32_t last_millis_{0};
-#endif
-  // millis_major_ is protected by lock when incrementing
+#endif /* else ESPHOME_CORES_MULTI_ATOMICS */
+
+  /*
+   * Upper 16 bits of the 64-bit millis counter. Incremented only while holding
+   * `lock_`; read concurrently. Atomic (relaxed) avoids a formal data race.
+   * Ordering relative to `last_millis_` is provided by its release store and the
+   * corresponding acquire loads.
+   */
+#ifdef ESPHOME_CORES_MULTI_ATOMICS
+  std::atomic<uint16_t> millis_major_{0};
+#else  /* not ESPHOME_CORES_MULTI_ATOMICS */
   uint16_t millis_major_{0};
-  uint32_t to_remove_{0};
+#endif /* else ESPHOME_CORES_MULTI_ATOMICS */
 };
 
 }  // namespace esphome
diff --git a/requirements.txt b/requirements.txt
index 4d94ce5557..6cc821e74c 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -12,7 +12,7 @@ platformio==6.1.18  # When updating platformio, also update /docker/Dockerfile
 esptool==4.9.0
 click==8.1.7
 esphome-dashboard==20250514.0
-aioesphomeapi==37.0.1
+aioesphomeapi==37.0.2
 zeroconf==0.147.0
 puremagic==1.30
 ruamel.yaml==0.18.14 # dashboard_import
diff --git a/script/api_protobuf/api_protobuf.py b/script/api_protobuf/api_protobuf.py
index 4df7692167..bb0e01d171 100755
--- a/script/api_protobuf/api_protobuf.py
+++ b/script/api_protobuf/api_protobuf.py
@@ -1491,6 +1491,28 @@ def find_common_fields(
     return common_fields
 
 
+def get_common_field_ifdef(
+    field_name: str, messages: list[descriptor.DescriptorProto]
+) -> str | None:
+    """Get the field_ifdef option if it's consistent across all messages.
+
+    Args:
+        field_name: Name of the field to check
+        messages: List of messages that contain this field
+
+    Returns:
+        The field_ifdef string if all messages have the same value, None otherwise
+    """
+    field_ifdefs = {
+        get_field_opt(field, pb.field_ifdef)
+        for msg in messages
+        if (field := next((f for f in msg.field if f.name == field_name), None))
+    }
+
+    # Return the ifdef only if all messages agree on the same value
+    return field_ifdefs.pop() if len(field_ifdefs) == 1 else None
+
+
 def build_base_class(
     base_class_name: str,
     common_fields: list[descriptor.FieldDescriptorProto],
@@ -1506,9 +1528,14 @@ def build_base_class(
     for field in common_fields:
         ti = create_field_type_info(field)
 
+        # Get field_ifdef if it's consistent across all messages
+        field_ifdef = get_common_field_ifdef(field.name, messages)
+
         # Only add field declarations, not encode/decode logic
-        protected_content.extend(ti.protected_content)
-        public_content.extend(ti.public_content)
+        if ti.protected_content:
+            protected_content.extend(wrap_with_ifdef(ti.protected_content, field_ifdef))
+        if ti.public_content:
+            public_content.extend(wrap_with_ifdef(ti.public_content, field_ifdef))
 
     # Determine if any message using this base class needs decoding
     needs_decode = any(