Skip to content

Kimi K2.5 can't enable thinking (reasoning_content missing #12090

Closed

Description

@3dot141

Description

What happened?

When using Kimi K2.5 via the OpenAI-compatible provider, enabling thinking does not return reasoning_content. As a result, the model behaves like non-thinking mode even when the feature should be available.

Expected behavior

Kimi K2.5 should support thinking and return interleaved reasoning content (reasoning_content) when enabled.

Steps to reproduce

  1. Configure a Kimi K2.5 model through the OpenAI-compatible provider.
  2. Enable thinking / reasoning in the request.
  3. Observe that no reasoning content is returned.

Environment

  • opencode version: v1.1.51
  • model: kimi-k2.5 / k2p5

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

Activity

  1. github-actions commented on Feb 4, 2026

    @github-actions
    Contributor

    This issue might be a duplicate of existing issues. Please check:

    Feel free to ignore if none of these address your specific case.

  2. leosongwei commented on Feb 4, 2026

    @leosongwei

    chould be this? vllm-project/vllm#33616

  3. docg1701 commented on Feb 4, 2026

    @docg1701

    It works fine with opencode-ai@1.1.49 but is broken since 1.1.50. I just checked. Kimi K2.5 got really dumb and non-thinking. Can't make it max/high variants with ctrl+t (which is missing past 1.1.49 in TUI).

    To fix this:
    add "autoupdate": false in ~/.config/opencode/opencode.json
    Then,
    npm install -g opencode-ai@1.1.49

    $ cat ~/.config/opencode/opencode.json 
    {
      "$schema": "https://opencode.ai/config.json",
        "tui": {
          "scroll_speed": 3,
          "scroll_acceleration": {
            "enabled": true
          }
       },
      "autoupdate": false,
      "mcp": {
        "context7": {
          "type": "local",
          "command": [
            "npx",
            "-y",
            "@upstash/context7-mcp",
            "--api-key",
            "ctx7sk-d860989a-2d42-48d6-9da8-4990ba4fd697"
          ],
          "enabled": true
        }
      }
    }
  4. rekram1-node commented on Feb 4, 2026

    @rekram1-node
    Contributor

    The kimi-for-coding provider has thinking off by default, I merged a fix to make it think by default in next release (consistent w/ other providers)

    The high/max actually makes no difference in thinking u just need to explicitly turn it on for that provider. I merged a "fix" from some external contributor who cited claims from kimi team directly.

    It caused this issue, but @monotykamary fixed it. you can fix temporarily using this:

     "provider": {
        "kimi-for-coding": {
          "models": {
            "k2p5": {
              "options": {
                 "thinking": {
                    "type": "enabled",
                    "budgetTokens": 16000 // this is required for ai sdk even tho it doesnt do anything in actuality
                  }
              }
            }
          }
        }
      }
    
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingzenRelates to Zen

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions