Skip to content

Kimi โ€‹

Updated

Endpoint: https://mmu.vod-qcloud.com/v1Authentication: VOD ApiToken / Bearer

Kimi is a text generation / chat model that also accepts image input (image_url). It is called through an OpenAI-compatible API, supporting the Chat Completions and Messages protocols with JSON / SSE streaming responses; k2.7-code targets coding scenarios.

Basic Information โ€‹

ItemValue
Call endpointhttps://mmu.vod-qcloud.com
AuthenticationVOD Application-issued ApiToken / Bearer
OutputText / Structured response
Response modeJSON / SSE

Supported Protocols โ€‹

SpecPathTarget
Chat CompletionsPOST /v1/chat/completionsAll models below
MessagesPOST /v1/messagesAll models below

The model IDs below use the corresponding spec. Authentication and common fields follow API Reference.

Sampling Options by Version โ€‹

Model IDtemperaturetop_p
kimi-k3ConfigurableConfigurable
kimi-k2.7-codeNot supportedNot supported
kimi-k2.6ConfigurableConfigurable

temperature specifies the sampling temperature, and top_p specifies the cumulative probability range. Adjust centered on one method.

Per-Version Reasoning Settings โ€‹

Model IDConfigurationDisable
kimi-k3thinking_enabledfalse
kimi-k2.7-codethinking_enabledCannot be disabled
kimi-k2.6thinking_enabledfalse

Based on Chat Completions. Reasoning content is read from reasoning_content, and the final answer from content. Do not pass a disable option to versions that cannot be disabled.

Request Parameters โ€‹

ParameterRequiredTypeDescription
modelRequiredStringA model ID from the per-version options table
messagesRequiredArrayConversation input for Chat Completions / Messages
max_tokensOptionalIntegerGeneration limit for Chat Completions / Messages. Required in Messages
streamOptionalBooleantrue: SSE / false: JSON

Request Examples โ€‹

Chat Completions โ€‹

json
{
  "model": "kimi-k3",
  "messages": [
    {
      "role": "user",
      "content": "Reply with exactly OK and nothing else."
    }
  ],
  "max_tokens": 2048,
  "stream": false,
  "reasoning_effort": "low"
}

Response โ€‹

json
{
  "id": "01a0de40-e125-7362-86bf-77c3082b129e",
  "object": "chat.completion",
  "model": "kimi-k3",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "OK",
        "reasoning_content": "The user wants exactly OK and nothing else. Final should be just OK. Need ensure no extra whitespace? Provide exactly OK."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 94,
    "prompt_tokens_details": {
      "audio_tokens": 0,
      "cached_tokens": 64
    },
    "prompt_cached_tokens_details": {
      "audio_tokens": 0
    },
    "completion_tokens": 39,
    "completion_tokens_details": {
      "accepted_prediction_tokens": 0,
      "audio_tokens": 0,
      "reasoning_tokens": 25,
      "rejected_prediction_tokens": 0
    },
    "total_tokens": 133
  }
}

Messages โ€‹

json
{
  "model": "kimi-k3",
  "messages": [
    {
      "role": "user",
      "content": "Reply with exactly OK and nothing else."
    }
  ],
  "max_tokens": 512,
  "stream": false
}

Response โ€‹

json
{
  "id": "01a0de34-c2f3-71f6-9a0f-32e6bfd963e1",
  "model": "kimi-k3",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "text": "OK",
      "type": "text"
    }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 93,
    "output_tokens": 146
  }
}

Special Settings โ€‹

Keep reasoning text separate from the final response, and check finish_reason along with it. Handle tool call responses with tool_calls, and preserve the tool ID and conversation order in follow-up requests. For image input, use the image_url Part of the corresponding version.

Built with VitePress ยท Deployed on Tencent EdgeOne Pages