anthropic, mistral: return model_name in response metadata

libs/partners/anthropic/langchain_anthropic/chat_models.py

@@ -900,6 +900,8 @@ def _format_output(self, data: Any, **kwargs: Any) -> ChatResult:
         llm_output = {
             k: v for k, v in data_dict.items() if k not in ("content", "role", "type")
         }
+        if "model" in llm_output and "model_name" not in llm_output:
+            llm_output["model_name"] = llm_output["model"]
         if (
             len(content) == 1
             and content[0]["type"] == "text"
@@ -1445,9 +1447,14 @@ def _make_message_chunk_from_anthropic_event(
     # See https://github.com/anthropics/anthropic-sdk-python/blob/main/src/anthropic/lib/streaming/_messages.py  # noqa: E501
     if event.type == "message_start" and stream_usage:
         usage_metadata = _create_usage_metadata(event.message.usage)
+        if hasattr(event.message, "model"):
+            response_metadata = {"model_name": event.message.model}
+        else:
+            response_metadata = {}
         message_chunk = AIMessageChunk(
             content="" if coerce_content_to_string else [],
             usage_metadata=usage_metadata,
+            response_metadata=response_metadata,
         )
     elif (
         event.type == "content_block_start"

libs/partners/anthropic/tests/integration_tests/test_chat_models.py

@@ -35,6 +35,7 @@ def test_stream() -> None:
     full: Optional[BaseMessageChunk] = None
     chunks_with_input_token_counts = 0
     chunks_with_output_token_counts = 0
+    chunks_with_model_name = 0
     for token in llm.stream("I'm Pickle Rick"):
         assert isinstance(token.content, str)
         full = token if full is None else full + token
@@ -44,12 +45,14 @@ def test_stream() -> None:
                 chunks_with_input_token_counts += 1
             elif token.usage_metadata.get("output_tokens"):
                 chunks_with_output_token_counts += 1
+        chunks_with_model_name += int("model_name" in token.response_metadata)
     if chunks_with_input_token_counts != 1 or chunks_with_output_token_counts != 1:
         raise AssertionError(
             "Expected exactly one chunk with input or output token counts. "
             "AIMessageChunk aggregation adds counts. Check that "
             "this is behaving properly."
         )
+    assert chunks_with_model_name == 1
     # check token usage is populated
     assert isinstance(full, AIMessageChunk)
     assert full.usage_metadata is not None
@@ -62,6 +65,7 @@ def test_stream() -> None:
     )
     assert "stop_reason" in full.response_metadata
     assert "stop_sequence" in full.response_metadata
+    assert "model_name" in full.response_metadata
 
 
 async def test_astream() -> None:
@@ -219,6 +223,7 @@ async def test_ainvoke() -> None:
 
     result = await llm.ainvoke("I'm Pickle Rick", config={"tags": ["foo"]})
     assert isinstance(result.content, str)
+    assert "model_name" in result.response_metadata
 
 
 def test_invoke() -> None:

libs/partners/mistralai/langchain_mistralai/chat_models.py

@@ -579,7 +579,11 @@ def _create_chat_result(self, response: Dict) -> ChatResult:
             )
             generations.append(gen)
 
-        llm_output = {"token_usage": token_usage, "model": self.model}
+        llm_output = {
+            "token_usage": token_usage,
+            "model_name": self.model,
+            "model": self.model,  # Backwards compatability
+        }
         return ChatResult(generations=generations, llm_output=llm_output)
 
     def _create_message_dicts(

libs/partners/mistralai/tests/integration_tests/test_chat_models.py

@@ -87,6 +87,7 @@ async def test_ainvoke() -> None:
 
     result = await llm.ainvoke("I'm Pickle Rick", config={"tags": ["foo"]})
     assert isinstance(result.content, str)
+    assert "model_name" in result.response_metadata
 
 
 def test_invoke() -> None: