How to Enable Thinking/Reasoning for Claude Models in Custom Endpoints?

[abhishekag03-cb]

Context:
I'm running LibreChat v0.7.9 with custom endpoints to access Claude models through my LLM gateway. I'm trying to enable the thinking/reasoning functionality for Claude models but can't get it to work through custom endpoints.

My Configuration:

version: 1.2.8
cache: true
  
endpoints:
  custom:
    # Anthropic Models via custom endpoint
    - name: "Anthropic "
      apiKey: "${MY_API_KEY}"
      baseURL: "https://<my-litellm-gateway-url>.net"
      models:
        default: [
          "claude-opus-4-1",
          "claude-sonnet-4",
          "claude-opus-4",
          "claude-3-7-sonnet",
          "claude-3-5-sonnet-v2",
          "claude-3-5-sonnet",
          "claude-3-5-haiku",
          "claude-3-haiku",
        ]
        fetch: false
      headers:
        X-Source: "LibreChat"
      titleConvo: true
      titleModel: "current_model"
      summarize: false
      summaryModel: "current_model"
      forcePrompt: false
      modelDisplayLabel: "Anthropic"
      iconURL: "anthropic"

What I Expected:

• Claude models (like claude-sonnet-4, claude-3-5-sonnet) should show thinking/reasoning controls in the UI settings
• When enabled, responses should include expandable thinking blocks showing the model's reasoning process

What I'm Seeing:

• No thinking/reasoning toggles appear in the custom endpoint settings UI
• Responses from Claude models don't include any thinking content, just regular responses
• The UI shows the same OpenAI-style parameters for all custom endpoints

Questions:

1. Is thinking/reasoning functionality supported for Claude models accessed through custom endpoints?
2. If yes, what configuration am I missing to enable it?
3. Do I need specific parameters in my librechat.yaml to enable thinking for custom Claude endpoints?
4. Should I be seeing thinking controls in the UI when using Claude models through custom endpoints?

Additional Info:

• When testing the same models through Anthropic's direct API, thinking works fine
• LibreChat version: 0.7.9
• Using Docker deployment

Any guidance on how to properly configure thinking for Claude models in custom endpoints would be greatly appreciated! 🙏

---

[danny-avila]

If you want reasoning always enabled, it's already possible and you need to make use of addParams.

If you want to have a custom parameters UI or use the pre-existing one for anthropic, and have these parameters behave exactly as anthropic API would expect them to be formatted, then this is planned to implement and is discussed here: #8933 (comment)

relevant links:

• https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint#addparams
• https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_params#picking-a-default-parameters-set