MODELS
Kimi K3
by Moonshot AI
Overview
Moonshot AI flagship 2.8T-parameter model with a 1M-token context window, native visual understanding, reasoning, long-horizon coding, and knowledge-work capabilities.
Details
Kimi K3 is Moonshot AI’s flagship model. Official Kimi API Platform sources list `kimi-k3` as Kimi’s most capable model, with 2.8T parameters, native visual understanding, and a 1M-token context window. Moonshot AI positions it for software engineering, knowledge work, deep reasoning, long-horizon coding, and agentic knowledge-work use cases. The official pricing page also documents API features including reasoning-effort support, automatic caching, tool calls, JSON mode, and structured output.
When to Use
Use Kimi K3 when you need Moonshot AI’s most capable documented Kimi model for long-context tasks up to a 1M-token context window. Use it for software engineering long-horizon coding knowledge work or deep-reasoning workflows where the official Kimi model catalog and pricing page identify Kimi K3 as the flagship option. Use it when native visual understanding or API features such as tool calls JSON mode structured output automatic caching or reasoning-effort support are required.
Getting Started
- Read the Kimi K3 launch page at https://www.kimi.com/blog/kimi-k3.
- Review the Kimi K3 quickstart documentation at https://platform.kimi.ai/docs/guide/kimi-k3-quickstart.
- Check the official model list for the `kimi-k3` model identifier and documented capabilities.
- Review the Kimi K3 pricing page before running API workloads.
Key Features
- •2.8T-parameter flagship model
- •1M-token context window
- •Native visual understanding
- •Reasoning and deep-reasoning use cases
- •Long-horizon coding and software-engineering use cases
- •Tool calls
- •JSON mode
- •and structured output documented for the API
- •Automatic caching and reasoning-effort support documented on the pricing page
Capabilities
- •text generation
- •long-context processing
- •vision
- •reasoning
- •coding
- •tool calling
- •JSON mode
- •structured output
- •agentic workflows
Last updated Jul 23, 2026