Kimi K3

Kimi K3 is Kimi’s most capable flagship model to date. With 2.8 trillion parameters, it is built on the Kimi Delta Attention (KDA) hybrid linear attention architecture and Attention Residuals technology. It natively supports visual understanding and features a 1-million-token context window. As the world’s first open-source model at the 3-trillion-parameter scale, Kimi K3 is designed for frontier AI use cases, including long-horizon coding, knowledge work, and reasoning.

Kimi K3

As the world’s first open-source model at the 3-trillion-parameter scale, Kimi K3 is designed for frontier AI use cases, including long-horizon coding, knowledge work, and reasoning.

Base URL

https://api.icreat.ai/llm/openai/v1

Authentication

All API requests must be authenticated with an API Key. You can obtain an API Key from the console.

export ICREAT_API_KEY="your-api-key-here"

HTTP Request Headers

import os

API_KEY = os.environ.get("ICREAT_API_KEY")
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer " + API_KEY,
}

Protect your API Key

Never expose your API Key in client-side code or public repositories. Use environment variables or a backend proxy.

Code Examples

This model is invoked via the OpenAI-compatible Chat Completions API and supports both streaming and non-streaming modes. With stream: false (default), the server returns the full JSON response at once. With stream: true, partial deltas are pushed as Server-Sent Events (SSE).

POST/chat/completions

curl --fail-with-body --connect-timeout 10 --max-time 60 \
  -X POST https://api.icreat.ai/llm/openai/v1/chat/completions \
  -H "Authorization: Bearer ${ICREAT_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "kimi-k3",
  "messages": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Coffee or tea ? Which is healthier? "}
  ],
  "thinking": {"type": "enabled"},
  "reasoning_effort": "low",
  "stream": false
}'

Input Schema

The following parameters are accepted in the request body.

Total: 6 Required: 2 Optional: 4

modelstringrequired

The model ID for the completion. Must be the iCreat model_code (kimi-k3).

Example: "kimi-k3"

messagesarray[object]required

Conversation messages.

max_tokensinteger

Maximum number of tokens to generate.

temperaturenumber

Sampling temperature, 0–2.

streamboolean

If true, stream via Server-Sent Events.

thinkingobject

Extended thinking configuration (if supported).

Output Schema

OpenAI-compatible Chat Completions response.

Total: 6

idstring

Unique completion identifier.

objectstring

Object type, always chat.completion.

createdinteger

Unix timestamp.

modelstring

Model ID used.

choicesarray[object]

List of completion choices.

usageobject

Token usage statistics.

LLM-friendly prompt

Below is an LLM-friendly Markdown prompt you can copy into Cursor, ChatGPT, or other AI assistants to help them understand this model's API integration, call flow, and key parameters.

# kimi-k3

> Kimi K3 is Kimi’s most capable flagship model to date.

## Overview

Call via iCreat OpenAI-compatible Chat Completions API; supports streaming and non-streaming and extended thinking.

## API Info

- **Base URL**: `https://api.icreat.ai/llm/openai/v1`
- **Endpoint (POST)**: `/chat/completions`
- **Model ID**: `kimi-k3`
- **Auth**: `Authorization: Bearer ${ICREAT_API_KEY}`

## Call Flow

Single POST; `stream: false` returns full JSON, `stream: true` streams via SSE.

### Input

- `model` (required): iCreat model_code `kimi-k3`
- `messages` (required): conversation messages
- Common optional: `max_tokens`, `temperature`, `stream`, `thinking`

### Output

- Read reply from `choices[0].message.content`

## Notes

- `model` must be the iCreat model_code
- Other fields follow the OpenAI Chat Completions protocol