Fix OpenAI prompt caching (#181472)

homeassistant/components/openai_conversation/entity.py

@@ -517,6 +517,7 @@ class OpenAIBaseLLMEntity(Entity):
             input=messages,
             max_output_tokens=options.get(CONF_MAX_TOKENS, RECOMMENDED_MAX_TOKENS),
             user=chat_log.conversation_id,
+            prompt_cache_key=self.subentry.subentry_id,
             service_tier=options.get(CONF_SERVICE_TIER, RECOMMENDED_SERVICE_TIER),
             store=options.get(CONF_STORE_RESPONSES, RECOMMENDED_STORE_RESPONSES),
             stream=True,

tests/components/openai_conversation/test_ai_task.py

@@ -66,6 +66,10 @@ async def test_generate_data(
     assert result.data == "The test data"
     assert mock_create_stream.call_args is not None
     assert mock_create_stream.call_args.kwargs["store"] is expected_store
+    assert (
+        mock_create_stream.call_args.kwargs["prompt_cache_key"]
+        == ai_task_entry.subentry_id
+    )
 
 
 @pytest.mark.usefixtures("mock_init_component")

tests/components/openai_conversation/test_conversation.py

@@ -823,13 +823,13 @@ async def test_flex_tier_retry(
 @pytest.mark.parametrize(
     "subentry_options", [{CONF_CHAT_MODEL: "gpt-5.6-sol", CONF_PRO_MODE: True}]
 )
+@pytest.mark.usefixtures("mock_init_component")
 async def test_model_args(
     hass: HomeAssistant,
     mock_config_entry: MockConfigEntry,
-    mock_init_component,
     mock_create_stream: AsyncMock,
     snapshot: SnapshotAssertion,
-    subentry_options: dict,
+    subentry_options: dict[str, str | bool],
 ) -> None:
     """Test model arguments for various configuration."""
 
@@ -860,4 +860,5 @@ async def test_model_args(
     model_args = mock_create_stream.call_args.kwargs.copy()
     model_args.pop("input")
     assert model_args.pop("user") == result.conversation_id
+    assert model_args.pop("prompt_cache_key") == subentry.subentry_id
     assert model_args == snapshot