Kimi K3.1 Pro - AnyFast

Kimi K3.1 Pro is Moonshot AI’s K3.1 flagship model for long-horizon coding, 3D creation, financial research, and end-to-end agent workflows. It is a 3-trillion-parameter-class model with native vision understanding and a 1,048,576-token context window. Kimi K3.1 Pro Highspeed uses the same model positioning and context window, with Moonshot AI describing output speed as approximately 2–3× the standard model. Choose the standard model for the default price/speed balance and Highspeed when response latency is the priority.

The public model IDs are kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Availability depends on the models enabled for your AnyFast project. This page follows the current Kimi API contract; confirm the IDs in your project’s model list before sending production traffic.

Model comparison

Model	Positioning	Context window	Speed	Reasoning
`kimi-k3.1-pro`	Standard K3.1 flagship for coding, 3D, finance, and long-horizon agents	1,048,576 tokens	Standard	Always on; `low`, `high`, or `max`
`kimi-k3.1-pro-highspeed`	High-speed variant with the same capability target	1,048,576 tokens	Approximately 2–3× the standard model, according to Moonshot AI	Always on; `low`, `high`, or `max`

Both models support text, image, and video input, function tools, structured output, streaming, Partial Mode, and automatic prefix caching when the unchanged prefix is eligible. Use the same request shape for both variants; only change the model value.

Quick example

curl https://www.anyfast.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3.1-pro",
    "reasoning_effort": "high",
    "messages": [
      {"role": "user", "content": "Explain how to review a large codebase safely."}
    ]
  }'

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://www.anyfast.ai/v1",
)

completion = client.chat.completions.create(
    model="kimi-k3.1-pro-highspeed",
    reasoning_effort="high",
    messages=[
        {"role": "user", "content": "Explain how to review a large codebase safely."}
    ],
)

print(completion.choices[0].message.content)

Reasoning effort

K3.1 Pro always reasons. Set the top-level reasoning_effort field to low, high, or max; the default is max. Do not send the K2.x thinking object to these models. Select the effort before a long conversation because changing it can prevent prefix-cache reuse.

completion = client.chat.completions.create(
    model="kimi-k3.1-pro",
    reasoning_effort="max",
    messages=[
        {"role": "user", "content": "Derive a robust migration plan for this service."}
    ],
)

Multimodal input

Send text, image, or video parts in a typed content array. AnyFast accepts public URLs and Base64 data URIs in the corresponding image_url or video_url part.

{
  "model": "kimi-k3.1-pro",
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "image_url", "image_url": {"url": "https://example.com/diagram.png"}},
        {"type": "text", "text": "Summarize this diagram."}
      ]
    }
  ]
}

Tools and multi-turn conversations

Define function tools at the top level and use tool_choice with auto, none, or required. After executing a tool, append the complete assistant message and the matching role: "tool" result for each tool_call_id before the next request. For any multi-turn request, preserve the complete assistant message returned by AnyFast, including reasoning_content and tool_calls. Do not reconstruct the assistant message from only its visible text.

Parameters and limits

Parameter	Type	Required	Description
`model`	string	Yes	`kimi-k3.1-pro` or `kimi-k3.1-pro-highspeed`
`messages`	array	Yes	Conversation history with text, image, video, or tool messages
`reasoning_effort`	string	No	`low`, `high`, or `max`; default `max`
`max_completion_tokens`	integer	No	Default 131,072; maximum 1,048,576. Input plus output must fit the 1M context window
`stream`	boolean	No	Enable SSE streaming; default `false`
`stream_options`	object	No	Set `include_usage: true` to receive usage in the final stream chunk
`response_format`	object	No	`text`, `json_object`, or `json_schema`
`tools`	array	No	Function definitions available to the model
`tool_choice`	string	No	`auto`, `none`, or `required`
`stop`	string / array	No	Up to five stop strings, each no longer than 32 bytes
`prompt_cache_key`	string	No	Stable session or task identifier for prefix-cache reuse
`safety_identifier`	string	No	Stable end-user identifier; prefer a one-way hash instead of direct personal data

Sampling parameters are fixed for the K3 series: temperature is 1.0, top_p is 0.95, n is 1, and presence_penalty and frequency_penalty are 0. Omit them from requests.

API reference

View the request and response schema for both K3.1 Pro model IDs.

Official Kimi documentation

Compare the provider’s current model guide, API examples, and limits.