Enhancement: Add support for DeepSeek reasoning model's content display

[RedwindA]

What features would you like to see added?

Support displaying reasoning content from DeepSeek's reasoning model responses in the UI, instead of only showing the final content.

More details

DeepSeek's reasoning model provides two types of content in its response:

• reasoning_content: Shows the model's step-by-step reasoning process
• content: Shows the final answer/conclusion

Currently, LibreChat only displays the 'content' part. To make better use of the reasoning model's capabilities, it would be valuable to show both the reasoning process and final content in the UI, helping users understand how the model arrived at its conclusions.

Reference: https://api-docs.deepseek.com/guides/reasoning_model

Which components are impacted by your request?

UI, Other

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Bold of them to not use OpenAI schema for this. I can look into, seems simple to add

[danny-avila]

#5379

[wbruntra]

I appreciate having the abililty to see the Chain of Thought, but on the other hand, it tends to be quite verbose compared to the final answer. Am I missing a setting to turn this display off/on? If not, it would be great to have such a setting.

With the new Deepseek model being so cheap I'm guessing it will see a lot of use.

[thomas-333]

I'm not seeing any Chain of Thought with Deepseek R1. I'm using the latest Docker image which includes the appropriate commit and I'm running Deepseek through OpenRouter if that makes a difference.

[thomas-333]

I believe that the Openrouter API has now been updated to include reasoning tokens:

https://openrouter.ai/docs/parameters#include-reasoning

I appreciate this will involve development work but it would be great if Librechat could be updated. Let me know if you need any testing carried out.

Edit: If this help this is an example of the full API output including reasoning:

{
  "id": "X",
  "provider": "DeepSeek",
  "model": "deepseek/deepseek-r1",
  "object": "chat.completion",
  "created": 1737716849,
  "choices": [
    {
      "logprobs": null,
      "finish_reason": "stop",
      "native_finish_reason": "stop",
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello! How can I assist you today?",
        "refusal": null,
        "reasoning": "Okay, the user sent \"Hello\". That's a friendly greeting. I should respond in a welcoming manner. Let me make sure to keep it open-ended so they feel comfortable to ask anything. Maybe something like, \"Hello! How can I assist you today?\" That should work."
      }
    }
  ],
  "system_fingerprint": "X",
  "usage": {
    "prompt_tokens": 6,
    "completion_tokens": 68,
    "total_tokens": 74
  }
}

[danny-avila]

Unfortunately this doesn't include the streaming tokens for reasoning

[thomas-333]

Apologies but I don't understand. Are you saying that the Openrouter API doesn't support streaming tokens? I didn't have streaming activated in the above API output, but if I do a small sample from my hacky Python code is:

[Reasoning Update]: Okay
[Reasoning Update]: ,
[Reasoning Update]:  the
[Reasoning Update]:  user
[Reasoning Update]:  just
[Reasoning Update]:  said
[Reasoning Update]:  "
[Reasoning Update]: Hello

[danny-avila]

Got it, on closer inspection I can see reasoning now, I was getting null reasoning tokens before

[thomas-333]

Thanks. I can confirm that this is now working for me. Cheers for adding this so quickly.