Kimi K2 Thinking
kimi-k2 series (incl. K2 Thinking, K2-0905, K2-0711) discontinued on the Kimi API on 2026-05-25; Moonshot recommends kimi-k3. Still available as open weights and via third parties. Pricing not verified.
- Context window
- 262,144 tokens
- Input
- text
- Output
- text
- License
- modified-mit
- Verified
- 2026-09-29
How to call it
| Provider | Model id | Endpoint / URL | Docs |
|---|---|---|---|
| Kimi API (discontinued) | kimi-k2-thinking / kimi-k2-thinking-turbo | https://api.moonshot.ai/v1 | docs |
| OpenRouter | moonshotai/kimi-k2-thinking | openrouter.ai/moonshotai/kimi-k2-thinking | — |
| Hugging Face | — | huggingface.co/moonshotai/Kimi-K2-Thinking | — |
Notable capabilities (3)
- Long tool-call chains: Interleaves reasoning with function calls and stays coherent across 200-300 sequential tool calls (vs 30-50 for earlier models, per Moonshot). source
- Native INT4 via quantization-aware training: QAT in post-training gives a lossless ~2x speed-up at INT4 on a 1T/32B-active MoE. source
- Heavy mode: Parallel 8-trajectory rollout with reflective aggregation used for top benchmark results (HLE, BrowseComp). source
Landmark open-weight thinking agent (Nov 2025). Use for self-hosting or research; for API use kimi-k3 or kimi-k2.6.
Sources: https://huggingface.co/moonshotai/Kimi-K2-Thinking · https://platform.kimi.ai/docs/models