Dropped from Moonshot's API in May, so agent builders can only use it by self-hosting the open weights.
Kenning's take
What it does
Kimi K2 Thinking is Moonshot AI's open-weight reasoning model, built on the same 1 trillion parameter design as Kimi K2 with a 256K context window. It mixes step-by-step reasoning with tool calls and, per Moonshot, stays stable across 200 to 300 calls in a row. Moonshot's API discontinued the K2 series on 25 May 2026, so it is now available as INT4 weights on Hugging Face.
- Interleaved reasoning and tool calls
- Stable across 200 to 300 sequential tool calls
- 256K token context
- Native INT4 quantisation
- Modified MIT licence
Who it suits. Developers building agents who want an open reasoning model they can host themselves.
Key facts
- Category
- AI models
- Part of
- Kimi
- Made by
- Moonshot AI
- Based in
- China
- Released
- 6 November 2025
- How to get it
- open weights on Hugging Face under a Modified MIT licence; Moonshot's API discontinued the K2 series on 25 May 2026
- Works on
- self-hosted, Hugging Face
- Free plan
- Weights free to download under the Modified MIT licence
Reviews and write-ups
- Simon Willison's Weblog 2025-11-06
FAQ
When was Kimi K2 Thinking released?
Early November 2025; Simon Willison wrote it up on 6 November 2025 and the Hugging Face repository dates from 4 November 2025.
Can I still use Kimi K2 Thinking?
Yes, by downloading the weights from Hugging Face. Moonshot's own API no longer supports it.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
Kimi K2
Moonshot AI's first trillion-parameter open-weight model, built for coding and tool use
Kimi K2.5
Moonshot AI's open multimodal model that added image and video input and agent swarms to Kimi
Kimi K2.6
Open Moonshot AI model for long coding sessions and swarms of up to 300 sub-agents
Kimi K2.7 Code
Moonshot AI's open coding model, built on K2.6 and using about 30% fewer thinking tokens