feat: Include model ID with token usage for live events

This allows users to track token usage data per model and fixes https://github.com/google/adk-python/issues/4084.

Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 853925212

tests/unittests/models/test_gemini_llm_connection.py

@@ -19,6 +19,8 @@ from google.adk.utils.variant_utils import GoogleLLMVariant
 from google.genai import types
 import pytest
 
+MODEL_VERSION = 'gemini-2.5-pro'
+
 
 @pytest.fixture
 def mock_gemini_session():
@@ -30,7 +32,9 @@ def mock_gemini_session():
 def gemini_connection(mock_gemini_session):
   """GeminiLlmConnection instance with mocked session."""
   return GeminiLlmConnection(
-      mock_gemini_session, api_backend=GoogleLLMVariant.VERTEX_AI
+      mock_gemini_session,
+      api_backend=GoogleLLMVariant.VERTEX_AI,
+      model_version=MODEL_VERSION,
   )
 
 
@@ -38,7 +42,9 @@ def gemini_connection(mock_gemini_session):
 def gemini_api_connection(mock_gemini_session):
   """GeminiLlmConnection instance with mocked session for Gemini API."""
   return GeminiLlmConnection(
-      mock_gemini_session, api_backend=GoogleLLMVariant.GEMINI_API
+      mock_gemini_session,
+      api_backend=GoogleLLMVariant.GEMINI_API,
+      model_version=MODEL_VERSION,
   )
 
 
@@ -215,6 +221,7 @@ async def test_receive_usage_metadata_and_server_content(
 
   usage_response = next((r for r in responses if r.usage_metadata), None)
   assert usage_response is not None
+  assert usage_response.model_version == MODEL_VERSION
   content_response = next((r for r in responses if r.content), None)
   assert content_response is not None
 

tests/unittests/models/test_google_llm.py

@@ -705,20 +705,27 @@ async def test_connect_without_custom_headers(gemini_llm, llm_request):
 
     mock_live_client.aio.live.connect.return_value = MockLiveConnect()
 
-    async with gemini_llm.connect(llm_request) as connection:
-      # Verify that the connect method was called with the right config
-      mock_live_client.aio.live.connect.assert_called_once()
-      call_args = mock_live_client.aio.live.connect.call_args
-      config_arg = call_args.kwargs["config"]
+    with mock.patch(
+        "google.adk.models.google_llm.GeminiLlmConnection"
+    ) as MockGeminiLlmConnection:
+      async with gemini_llm.connect(llm_request) as connection:
+        # Verify that the connect method was called with the right config
+        mock_live_client.aio.live.connect.assert_called_once()
+        call_args = mock_live_client.aio.live.connect.call_args
+        config_arg = call_args.kwargs["config"]
 
-      # Verify that http_options remains None since no custom headers were provided
-      assert config_arg.http_options is None
+        # Verify that http_options remains None since no custom headers were provided
+        assert config_arg.http_options is None
 
-      # Verify that system instruction and tools were still set
-      assert config_arg.system_instruction is not None
-      assert config_arg.tools == llm_request.config.tools
+        # Verify that system instruction and tools were still set
+        assert config_arg.system_instruction is not None
+        assert config_arg.tools == llm_request.config.tools
 
-      assert isinstance(connection, GeminiLlmConnection)
+        MockGeminiLlmConnection.assert_called_once_with(
+            mock_live_session,
+            api_backend=gemini_llm._api_backend,
+            model_version=llm_request.model,
+        )
 
 
 @pytest.mark.parametrize(