fix(models): populate finish_reason on AnthropicLlm responses

Merge #5512

## Summary

Fixes #5394

`AnthropicLlm` was calling `to_google_genai_finish_reason()` but never wiring
its result into `LlmResponse`, so `finish_reason` was always `None`. This fix
connects it end-to-end and expands the mapping to cover all stop reasons the
current Anthropic API can return.

- **`to_google_genai_finish_reason`** — extended to cover `pause_turn` → `STOP`
  (streaming interruption, e.g. server-side tool use) and `refusal` → `SAFETY`
  (model refusal); all other unknown values continue to return
  `FINISH_REASON_UNSPECIFIED`
- **`message_to_generate_content_response`** — passes `stop_reason` through
  `to_google_genai_finish_reason` and sets it on the returned `LlmResponse`
- **`_generate_content_streaming`** — reads `stop_reason` from the
  `message_delta` event and sets `finish_reason` on the final aggregated
  `LlmResponse`

Co-authored-by: Jason Zhang <jasoncz@google.com>
PiperOrigin-RevId: 952921402

Author: 2 people

src/google/adk/models/anthropic_llm.py

@@ -231,14 +231,26 @@ def to_claude_role(role: Optional[str]) -> Literal["user", "assistant"]:
   return "user"
 
 
+# Mapping of Anthropic stop_reason strings to FinishReason enum values
+_STOP_REASON_MAPPING: dict[anthropic_types.StopReason, types.FinishReason] = {
+    "end_turn": types.FinishReason.STOP,
+    "stop_sequence": types.FinishReason.STOP,
+    "tool_use": types.FinishReason.STOP,
+    "pause_turn": types.FinishReason.STOP,
+    "max_tokens": types.FinishReason.MAX_TOKENS,
+    "refusal": types.FinishReason.SAFETY,
+}
+
+
 def to_google_genai_finish_reason(
-    anthropic_stop_reason: Optional[str],
-) -> types.FinishReason:
-  if anthropic_stop_reason in ["end_turn", "stop_sequence", "tool_use"]:
-    return "STOP"
-  if anthropic_stop_reason == "max_tokens":
-    return "MAX_TOKENS"
-  return "FINISH_REASON_UNSPECIFIED"
+    anthropic_stop_reason: Optional[anthropic_types.StopReason],
+) -> types.FinishReason | None:
+  """Maps Anthropic stop_reason to Google GenAI FinishReason."""
+  if anthropic_stop_reason is None:
+    return None
+  return _STOP_REASON_MAPPING.get(
+      anthropic_stop_reason, types.FinishReason.FINISH_REASON_UNSPECIFIED
+  )
 
 
 def _is_image_part(part: types.Part) -> bool:
@@ -499,8 +511,7 @@ def message_to_generate_content_response(
           ),
           cached_content_token_count=_extract_cached_token_count(message.usage),
       ),
-      # TODO: Deal with these later.
-      # finish_reason=to_google_genai_finish_reason(message.stop_reason),
+      finish_reason=to_google_genai_finish_reason(message.stop_reason),
   )
 
 
@@ -788,6 +799,7 @@ async def _generate_content_streaming(
     input_tokens = 0
     output_tokens = 0
     cached_input_tokens: int | None = None
+    stop_reason: Optional[anthropic_types.StopReason] = None
 
     async for event in raw_stream:
       if event.type == "message_start":
@@ -859,6 +871,7 @@ async def _generate_content_streaming(
 
       elif event.type == "message_delta":
         output_tokens = event.usage.output_tokens
+        stop_reason = event.delta.stop_reason
 
     # Build the final aggregated response with all content.
     all_parts: list[types.Part] = []
@@ -901,6 +914,7 @@ async def _generate_content_streaming(
             total_token_count=input_tokens + output_tokens,
             cached_content_token_count=cached_input_tokens,
         ),
+        finish_reason=to_google_genai_finish_reason(stop_reason),
         partial=False,
     )
 

tests/unittests/models/test_anthropic_llm.py

@@ -31,6 +31,7 @@
 from google.adk.models.anthropic_llm import content_to_message_param
 from google.adk.models.anthropic_llm import function_declaration_to_tool_param
 from google.adk.models.anthropic_llm import part_to_message_block
+from google.adk.models.anthropic_llm import to_google_genai_finish_reason
 from google.adk.models.llm_request import LlmRequest
 from google.adk.models.llm_response import LlmResponse
 from google.genai import types
@@ -2642,3 +2643,121 @@ async def mock_coro():
       assert "top_p" not in kwargs
       assert "top_k" not in kwargs
       assert kwargs["output_config"] == {"effort": "xhigh"}
+
+
+@pytest.mark.parametrize(
+    "stop_reason,expected_finish_reason",
+    [
+        ("end_turn", types.FinishReason.STOP),
+        ("stop_sequence", types.FinishReason.STOP),
+        ("tool_use", types.FinishReason.STOP),
+        ("max_tokens", types.FinishReason.MAX_TOKENS),
+        ("pause_turn", types.FinishReason.STOP),
+        ("refusal", types.FinishReason.SAFETY),
+        (None, None),
+        ("unknown", types.FinishReason.FINISH_REASON_UNSPECIFIED),
+    ],
+)
+def test_to_google_genai_finish_reason(stop_reason, expected_finish_reason):
+  """All Anthropic stop_reason values map to the correct ADK FinishReason."""
+  assert to_google_genai_finish_reason(stop_reason) == expected_finish_reason
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "stop_reason,expected_finish_reason",
+    [
+        ("end_turn", types.FinishReason.STOP),
+        ("max_tokens", types.FinishReason.MAX_TOKENS),
+        ("refusal", types.FinishReason.SAFETY),
+    ],
+)
+async def test_non_streaming_sets_finish_reason(
+    stop_reason, expected_finish_reason
+):
+  """finish_reason is populated on the non-streaming LlmResponse."""
+  llm = AnthropicLlm(model="claude-sonnet-4-20250514")
+  mock_message = anthropic_types.Message(
+      id="msg_test",
+      content=[
+          anthropic_types.TextBlock(text="Hi", type="text", citations=None)
+      ],
+      model="claude-sonnet-4-20250514",
+      role="assistant",
+      stop_reason=stop_reason,
+      stop_sequence=None,
+      type="message",
+      usage=anthropic_types.Usage(
+          input_tokens=5,
+          output_tokens=2,
+          cache_creation_input_tokens=0,
+          cache_read_input_tokens=0,
+          server_tool_use=None,
+          service_tier=None,
+      ),
+  )
+  mock_client = MagicMock()
+  mock_client.messages.create = AsyncMock(return_value=mock_message)
+
+  llm_request = LlmRequest(
+      model="claude-sonnet-4-20250514",
+      contents=[Content(role="user", parts=[Part.from_text(text="Hi")])],
+      config=types.GenerateContentConfig(system_instruction="Test"),
+  )
+
+  with mock.patch.object(llm, "_anthropic_client", mock_client):
+    responses = [
+        r async for r in llm.generate_content_async(llm_request, stream=False)
+    ]
+
+  assert len(responses) == 1
+  assert responses[0].finish_reason == expected_finish_reason
+
+
+@pytest.mark.asyncio
+async def test_streaming_sets_finish_reason():
+  """finish_reason is populated on the final streaming LlmResponse."""
+  llm = AnthropicLlm(model="claude-sonnet-4-20250514")
+
+  events = [
+      MagicMock(
+          type="message_start",
+          message=MagicMock(usage=MagicMock(input_tokens=5, output_tokens=0)),
+      ),
+      MagicMock(
+          type="content_block_start",
+          index=0,
+          content_block=anthropic_types.TextBlock(text="", type="text"),
+      ),
+      MagicMock(
+          type="content_block_delta",
+          index=0,
+          delta=anthropic_types.TextDelta(text="Hi", type="text_delta"),
+      ),
+      MagicMock(type="content_block_stop", index=0),
+      MagicMock(
+          type="message_delta",
+          delta=MagicMock(stop_reason="max_tokens"),
+          usage=MagicMock(output_tokens=1),
+      ),
+      MagicMock(type="message_stop"),
+  ]
+
+  mock_client = MagicMock()
+  mock_client.messages.create = AsyncMock(
+      return_value=_make_mock_stream_events(events)
+  )
+
+  llm_request = LlmRequest(
+      model="claude-sonnet-4-20250514",
+      contents=[Content(role="user", parts=[Part.from_text(text="Hi")])],
+      config=types.GenerateContentConfig(system_instruction="Test"),
+  )
+
+  with mock.patch.object(llm, "_anthropic_client", mock_client):
+    responses = [
+        r async for r in llm.generate_content_async(llm_request, stream=True)
+    ]
+
+  final = responses[-1]
+  assert final.finish_reason == types.FinishReason.MAX_TOKENS