Repository navigation
Kimi models in OpenCode Go ignore the "thinking": {"type": "disabled"} parameter #25001
Copy link
Copy link
Closed as not planned
Closed as not planned
Copy link
Description
Description
I'm maintaining a VSCode plugin (opencode-go-copilot) and received user reports that Kimi's non-thinking mode does not work. After testing, I found that Kimi's model does not respond to the "thinking": {"type": "disabled"} parameter at all — it always keeps thinking enabled.
Below are the raw request and response from my test:
Details
{
"model": "kimi-k2.6",
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "Please introduce Kimi models in one sentence."
}
],
"temperature": 0.7,
"max_tokens": 2048,
"thinking": {
"type": "disabled"
}
}
{
"id": "gen-1777491101-1pYb8Qpou2w58yiX4lxd",
"object": "chat.completion",
"created": 1777491101,
"model": "moonshotai/kimi-k2.6-20260420",
"provider": "Moonshot AI",
"system_fingerprint": "fpv0_8cab9aba",
"choices": [
{
"index": 0,
"logprobs": null,
"finish_reason": "stop",
"native_finish_reason": "stop",
"message": {
"role": "assistant",
"content": "Kimi models are a family of large language models developed by the Chinese AI company Moonshot AI, distinguished by their ability to process extremely long contexts and perform complex multimodal tasks.",
"refusal": null,
"reasoning": "The user wants a one-sentence introduction to \"Kimi models\". This likely refers to the Kimi series of large language models developed by Moonshot AI (月之暗面), a Chinese AI company. The Kimi models include Kimi Chat, Kimi K1.5, etc., known for their long context windows (up to 2 million characters in some versions).\n\nI need to provide a concise, accurate one-sentence introduction. Key points to include:\n- Developed by Moonshot AI\n- Large language models/AI assistant\n- Known for long context window capabilities\n- Chinese origin but capable in multiple languages\n\nLet me craft a single sentence that captures the essence:\n\n\"Kimi models are a series of large language models developed by Chinese AI company Moonshot AI, renowned for their exceptionally long context windows and strong multimodal reasoning capabilities.\"\n\nOr more simply:\n\"Kimi is a family of large language models created by Moonshot AI that specializes in processing extremely long contexts and engaging in multilingual conversations.\"\n\nI should make sure it's factual and neutral. The user asked for \"one sentence\" so I must strictly keep it to one sentence.\n\nLet me check: Kimi models are indeed by Moonshot AI (月之暗面). They are known for long context (2M tokens/characters). They have versions like Kimi-v1, Kimi Chat, Kimi K1.5 (with reasoning). \n\nBest one-sentence version:\n\"Kimi models are a family of large language models developed by the Chinese AI company Moonshot AI, distinguished by their ability to process extremely long contexts and perform complex multimodal tasks.\"\n\nThis is accurate, concise, and one sentence.",
"reasoning_details": [
{
"type": "reasoning.text",
"text": "The user wants a one-sentence introduction to \"Kimi models\". This likely refers to the Kimi series of large language models developed by Moonshot AI (月之暗面), a Chinese AI company. The Kimi models include Kimi Chat, Kimi K1.5, etc., known for their long context windows (up to 2 million characters in some versions).\n\nI need to provide a concise, accurate one-sentence introduction. Key points to include:\n- Developed by Moonshot AI\n- Large language models/AI assistant\n- Known for long context window capabilities\n- Chinese origin but capable in multiple languages\n\nLet me craft a single sentence that captures the essence:\n\n\"Kimi models are a series of large language models developed by Chinese AI company Moonshot AI, renowned for their exceptionally long context windows and strong multimodal reasoning capabilities.\"\n\nOr more simply:\n\"Kimi is a family of large language models created by Moonshot AI that specializes in processing extremely long contexts and engaging in multilingual conversations.\"\n\nI should make sure it's factual and neutral. The user asked for \"one sentence\" so I must strictly keep it to one sentence.\n\nLet me check: Kimi models are indeed by Moonshot AI (月之暗面). They are known for long context (2M tokens/characters). They have versions like Kimi-v1, Kimi Chat, Kimi K1.5 (with reasoning). \n\nBest one-sentence version:\n\"Kimi models are a family of large language models developed by the Chinese AI company Moonshot AI, distinguished by their ability to process extremely long contexts and perform complex multimodal tasks.\"\n\nThis is accurate, concise, and one sentence.",
"format": "unknown",
"index": 0
}
]
}
}
],
"usage": {
"prompt_tokens": 26,
"completion_tokens": 379,
"total_tokens": 405,
"cost": 0,
"is_byok": true,
"prompt_tokens_details": {
"cached_tokens": 0,
"cache_write_tokens": 0,
"audio_tokens": 0,
"video_tokens": 0
},
"cost_details": {
"upstream_inference_cost": 0.0015407,
"upstream_inference_prompt_cost": 2.47e-05,
"upstream_inference_completions_cost": 0.001516
},
"completion_tokens_details": {
"reasoning_tokens": 422,
"image_tokens": 0,
"audio_tokens": 0
}
},
"cost": "0"
}
Plugins
No response
OpenCode version
No response
Steps to reproduce
- Use the OpenAI SDK to call OpenCode Go's Kimi-K2.6 (or Kimi-K2.5) model, and include the
thinkingparameter as{"type": "disabled"}in the request. - Retrieve the
choicefrom the response, and observe thatmessage.reasoningstill contains the thinking chain (i.e., reasoning output is not disabled).
Screenshot and/or share link
No response
Operating System
No response
Terminal
No response
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels
Activity
[-]Kimi model does not respect thinking: disable parameter[/-][+]Kimi models in OpenCode Go ignore the "thinking": {"type": "disable"} parameter[/+][-]Kimi models in OpenCode Go ignore the "thinking": {"type": "disable"} parameter[/-][+]Kimi models in OpenCode Go ignore the "thinking": {"type": "disabled"} parameter[/+]faisstzera commentedon Apr 29, 2026
Same for deepseek models, impossible to disable reasoning.
OnesoftQwQ commentedon May 4, 2026
But DeepSeek models can disable reasoning in my test. Maybe your parameter is incorrect?
OnesoftQwQ commentedon May 4, 2026
Also, GLM models can't disable reasoning either.
github-actions commentedon Jul 5, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.