DeepSeek V3 Does Not Support Structured Output in LangChain with ChatOpenAI()

[ksmooi]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

Steps to Reproduce:

1. Install the required libraries:

   !pip install -qU langchain-openai
   !pip install -qU langchain_community
   !pip install -qU langchain_experimental
   !pip install -qU langgraph

2. Initialize the model and define a Pydantic model for structured output:

   from langchain_openai import ChatOpenAI
   from kaggle_secrets import UserSecretsClient
   from pydantic import BaseModel, Field
   
   llm_api_key = UserSecretsClient().get_secret("api-key-deepseek")
   model = ChatOpenAI(model="deepseek-chat", temperature=0, openai_api_key=llm_api_key, openai_api_base='https://api.deepseek.com')
   
   # Define a Pydantic model for structured output
   class Person(BaseModel):
       name: str = Field(description="The name of the person")
       age: int = Field(description="The age of the person")
       email: str = Field(description="The email address of the person")
   
   # Query the model (correct)
   response = model.invoke("Extract the name, age, and email of John Doe, who is 30 years old and has the email [email protected].")
   print(response)

3. Use with_structured_output() to enforce the Pydantic model and query the model with structured output:

   structured_model = model.with_structured_output(Person)
   
   # Query the model with structured output (incorrect)
   response = structured_model.invoke("Extract the name, age, and email of John Doe, who is 30 years old and has the email [email protected].")
   print(response)

Error Message and Stack Trace (if applicable)

Actual Behavior:

The model throws an UnprocessableEntityError indicating that the response_format type json_schema is unavailable:

content='Here is the extracted information:\n\n- **Name**: John Doe  \n- **Age**: 30  \n- **Email**: [email protected]' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 30, 'prompt_tokens': 33, 'total_tokens': 63, 'completion_tokens_details': None, 'prompt_tokens_details': None, 'prompt_cache_hit_tokens': 0, 'prompt_cache_miss_tokens': 33}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_3a5770e1b4', 'finish_reason': 'stop', 'logprobs': None} id='run-d078cad9-42a0-4be0-9e92-a593002a8606-0' usage_metadata={'input_tokens': 33, 'output_tokens': 30, 'total_tokens': 63, 'input_token_details': {}, 'output_token_details': {}}

---------------------------------------------------------------------------
UnprocessableEntityError                  Traceback (most recent call last)
<ipython-input-14-83b3c0097ccc> in <cell line: 18>()
     16 
     17 # Query the model with structured output
---> 18 response = structured_model.invoke("Extract the name, age, and email of John Doe, who is 30 years old and has the email [email protected].")
     19 print(response)

/usr/local/lib/python3.10/dist-packages/langchain_core/runnables/base.py in invoke(self, input, config, **kwargs)
   3018                 context.run(_set_config_context, config)
   3019                 if i == 0:
-> 3020                     input = context.run(step.invoke, input, config, **kwargs)
   3021                 else:
   3022                     input = context.run(step.invoke, input, config)

/usr/local/lib/python3.10/dist-packages/langchain_core/runnables/base.py in invoke(self, input, config, **kwargs)
   5350         **kwargs: Optional[Any],
   5351     ) -> Output:
-> 5352         return self.bound.invoke(
   5353             input,
   5354             self._merge_configs(config),

/usr/local/lib/python3.10/dist-packages/langchain_core/language_models/chat_models.py in invoke(self, input, config, stop, **kwargs)
    284         return cast(
    285             ChatGeneration,
--> 286             self.generate_prompt(
    287                 [self._convert_input(input)],
    288                 stop=stop,

/usr/local/lib/python3.10/dist-packages/langchain_core/language_models/chat_models.py in generate_prompt(self, prompts, stop, callbacks, **kwargs)
    784     ) -> LLMResult:
    785         prompt_messages = [p.to_messages() for p in prompts]
--> 786         return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)
    787 
    788     async def agenerate_prompt(

/usr/local/lib/python3.10/dist-packages/langchain_core/language_models/chat_models.py in generate(self, messages, stop, callbacks, tags, metadata, run_name, run_id, **kwargs)
    641                 if run_managers:
    642                     run_managers[i].on_llm_error(e, response=LLMResult(generations=[]))
--> 643                 raise e
    644         flattened_outputs = [
    645             LLMResult(generations=[res.generations], llm_output=res.llm_output)  # type: ignore[list-item]

/usr/local/lib/python3.10/dist-packages/langchain_core/language_models/chat_models.py in generate(self, messages, stop, callbacks, tags, metadata, run_name, run_id, **kwargs)
    631             try:
    632                 results.append(
--> 633                     self._generate_with_cache(
    634                         m,
    635                         stop=stop,

/usr/local/lib/python3.10/dist-packages/langchain_core/language_models/chat_models.py in _generate_with_cache(self, messages, stop, run_manager, **kwargs)
    849         else:
    850             if inspect.signature(self._generate).parameters.get("run_manager"):
--> 851                 result = self._generate(
    852                     messages, stop=stop, run_manager=run_manager, **kwargs
    853                 )

/usr/local/lib/python3.10/dist-packages/langchain_openai/chat_models/base.py in _generate(self, messages, stop, run_manager, **kwargs)
    771             payload.pop("stream")
    772             try:
--> 773                 response = self.root_client.beta.chat.completions.parse(**payload)
    774             except openai.BadRequestError as e:
    775                 _handle_openai_bad_request(e)

/usr/local/lib/python3.10/dist-packages/openai/resources/beta/chat/completions.py in parse(self, messages, model, audio, response_format, frequency_penalty, function_call, functions, logit_bias, logprobs, max_completion_tokens, max_tokens, metadata, modalities, n, parallel_tool_calls, prediction, presence_penalty, reasoning_effort, seed, service_tier, stop, store, stream_options, temperature, tool_choice, tools, top_logprobs, top_p, user, extra_headers, extra_query, extra_body, timeout)
    158             )
    159 
--> 160         return self._post(
    161             "/chat/completions",
    162             body=maybe_transform(

/usr/local/lib/python3.10/dist-packages/openai/_base_client.py in post(self, path, cast_to, body, options, files, stream, stream_cls)
   1281             method="post", url=path, json_data=body, files=to_httpx_files(files), **options
   1282         )
-> 1283         return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))
   1284 
   1285     def patch(

/usr/local/lib/python3.10/dist-packages/openai/_base_client.py in request(self, cast_to, options, remaining_retries, stream, stream_cls)
    958             retries_taken = 0
    959 
--> 960         return self._request(
    961             cast_to=cast_to,
    962             options=options,

/usr/local/lib/python3.10/dist-packages/openai/_base_client.py in _request(self, cast_to, options, retries_taken, stream, stream_cls)
   1062 
   1063             log.debug("Re-raising status error")
-> 1064             raise self._make_status_error_from_response(err.response) from None
   1065 
   1066         return self._process_response(

UnprocessableEntityError: Failed to deserialize the JSON body into the target type: response_format: response_format.type `json_schema` is unavailable now at line 1 column 626

Description

Description:

When using ChatOpenAI() with DeepSeek V3 in LangChain, the with_structured_output() method fails to enforce structured output formats (e.g., Pydantic models). The model returns an error indicating that the response_format type json_schema is unavailable. This prevents the use of structured output functionality, which is critical for applications requiring consistent and predictable data formats.

Expected Behavior:

The model should return a structured output in the format defined by the Pydantic model:

Person(name="John Doe", age=30, email="[email protected]")

System Info

Environment:

• Python 3.x (Kaggle Notebook)
• Libraries: langchain-openai, langchain-community, langchain_experimental, langgraph
• Model: deepseek-chat (via DeepSeek API)