Skip to main content
Kimi-K2.5 is a multimodal chat model by MoonShot (月之暗面), available through Anyfast via an OpenAI-compatible interface. It achieves open-source SoTA across agent, coding, and vision-understanding tasks, and supports vision and text input, thinking and non-thinking modes, and both conversational and agent workloads.

​
Key capabilities

  • OpenAI-compatible — Works as a drop-in replacement with the OpenAI SDK
  • 256K context — 262,144 tokens for large documents and multi-turn conversations
  • Vision understanding — Accepts image and text input; reads text, colors, and shapes in images
  • Thinking mode — Toggle via the thinking parameter; returns reasoning_content
  • Agent & coding — Open-source SoTA on agent and code tasks
  • Rich features — Tool Calls (function calling), JSON Mode, Partial Mode, web search, and automatic context caching

​
Quick example

curl https://www.anyfast.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k2.5",
    "messages": [
      { "role": "user", "content": "Explain quantum entanglement in simple terms." }
    ]
  }'
Note: For vision, message.content must be an array of objects (not a string), and do not serialize the array into a string. image_url accepts a base64 data URI (data:image/png;base64,...) or a file reference (ms://<file_id>). Tokens served from the context cache are reported in usage.prompt_tokens_details.cached_tokens.

​
Parameters

ParameterTypeRequiredDescription
modelstringYesMust be kimi-k2.5
messagesarrayYesList of { role, content } objects. content may be a string or a multimodal array of text/image_url
thinkingobjectNoControls thinking mode, e.g. {"type": "enabled"} (default) or {"type": "disabled"}
max_completion_tokensintegerNoMaximum tokens to generate. (max_tokens is deprecated and not honored)
temperaturefloatNo0–2. Controls randomness. Default: 1
streambooleanNoEnable SSE streaming. Default: false
top_pfloatNoNucleus sampling threshold. Default: 1
response_formatobjectNoSet to {"type": "json_object"} to enable JSON Mode
toolsarrayNoA list of tools the model may call (function calling)
stopstring / arrayNoSequences that stop generation (up to 5, max 32 bytes each)

API Reference

View the interactive API playground for Kimi-K2.5.