ChatOllama: AIMessageChunk missing reasoning_content in additional_kwargs (concatenated into content instead)

[djholt]

Checked other resources

• This is a bug, not a usage question. For questions, please use the LangChain Forum (https://forum.langchain.com/).
• I added a very descriptive title to this issue.
• I searched the LangChain.js 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.js 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

const llm = new ChatOllama({ 
  baseUrl: 'http://localhosr:11434',
  model: 'gpt-oss:20b',
  think: false // or true, same issue
});

const message = new HumanMessage({
  content: [
    {
      type: 'text',
      text: 'Hello'
    }
  ]
});

const response = await llm.invoke([message]);
// response.content includes reasoning/thinking from gpt-oss:20b

Error Message and Stack Trace (if applicable)

No response

Description

Looking at convertOllamaMessagesToLangChain, I can see that thinking and content are merely concatenated. This differs from langchain Python wherein thinking is returned as reasoning_content in additional_kwargs. This is the expected behavior.

System Info

"@langchain/core": "^0.3.78",
"@langchain/langgraph": "^0.4.9",
"@langchain/ollama": "^0.2.4",

[pawel-twardziak]

Hi @djholt thaks for reporting this issue.
Could you please provide two examples: response from python lib and response from js lib?

[djholt]

Absolutely. Here are output examples for both. Thank you!

Python with reasoning=True (expected behavior):

content='Hello! How can I help you today?' additional_kwargs={'reasoning_content': 'The user says "Hello". Probably they want a greeting. We should respond politely.'} response_metadata={'model': 'gpt-oss:20b', 'created_at': '2025-10-02T14:47:39.275402Z', 'done': True, 'done_reason': 'stop', 'total_duration': 493927375, 'load_duration': 106353667, 'prompt_eval_count': 68, 'prompt_eval_duration': 22218833, 'eval_count': 36, 'eval_duration': 364866875, 'model_name': 'gpt-oss:20b'} id='run--d9fa1fe1-2d3f-40b6-bfe7-ca7dbb3b04f9-0' usage_metadata={'input_tokens': 68, 'output_tokens': 36, 'total_tokens': 104}

JavaScript with think: true (or think: false—the behavior is the same):

AIMessage {
  "content": "We need to respond to greeting. Should be friendly. Probably ask how to help.Hello! 👋 How can I assist you today?",
  "additional_kwargs": {},
  "response_metadata": {
    "model": "gpt-oss:20b",
    "created_at": "2025-10-02T14:51:30.195675Z",
    "done": true,
    "done_reason": "stop",
    "total_duration": 521472000,
    "load_duration": 107368292,
    "prompt_eval_count": 68,
    "prompt_eval_duration": 24266042,
    "eval_count": 38,
    "eval_duration": 389455833
  },
  "tool_calls": [],
  "invalid_tool_calls": [],
  "usage_metadata": {
    "input_tokens": 68,
    "output_tokens": 38,
    "total_tokens": 106
  }
}

Expected JavaScript behavior:

AIMessage {
  "content": "Hello! 👋 How can I assist you today?",
  "additional_kwargs": {
    "reasoning_content": "We need to respond. It's a greeting. Just a friendly greeting. Probably ask how can I help."
  },
  "response_metadata": {
    "model": "gpt-oss:20b",
    "created_at": "2025-10-02T14:52:57.897298Z",
    "done": true,
    "done_reason": "stop",
    "total_duration": 550552958,
    "load_duration": 109742625,
    "prompt_eval_count": 68,
    "prompt_eval_duration": 21026000,
    "eval_count": 42,
    "eval_duration": 419394500
  },
  "tool_calls": [],
  "invalid_tool_calls": [],
  "usage_metadata": {
    "input_tokens": 68,
    "output_tokens": 42,
    "total_tokens": 110
  }
}

[pawel-twardziak]

@djholt thanks! Let me investigate further on this matter

[pawel-twardziak]

hi @djholt I've just pushed my PR. Let's see whether this gets accepted and merged.
What are your thoughts about this solution?

[djholt]

@pawel-twardziak Thank you! Yes, this looks great. My temporary fix was very similar.

My only other thought would be to consider the think: true argument, but I'm not exactly sure what the behavior difference should be.

I appreciate you taking the time to work on this! 🙂

[pawel-twardziak]

Hi @pawel-twardziak your welcome 🥲

btw, what do you mean by think: true argument?

[djholt]

See here. It's not documented very well, but I believe it's intended to be similar to Python's reasoning constructor parameter. Go to this page and take a look at the section titled "Thinking / Reasoning".

[pawel-twardziak]

Yes, it would be nice to have them aligned. But I think I won't be part of the scope for this bug-fix :)

[djholt]

And that's fair enough. Maybe a future feature/bug issue. Thanks again!