Moonshot AI releases the weights of Kimi K3.1 Pro, a 2.8T-parameter open model with a 1M-token context
The mixture-of-experts model activates 104B parameters per token and ships under a custom licence that requires large commercial users to display its name; Moonshot's own benchmarks put it just behind Claude Opus 5.5 on Terminal-Bench 4.0.
Why it matters
Kimi K3.1 Pro went on sale through Moonshot's apps and API first, with the company promising the full weights by 4 November, and it has now kept that promise. Moonshot reports 65.7% on Terminal-Bench 4.0, ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%, so an openly downloadable model now sits close to the best closed systems on agentic coding. The licence asks products with over 100 million monthly users or $20 million in monthly revenue to show the model's name. Watch for independent benchmark checks and how quickly hosting providers adopt it.
Key facts
- 2.8T total parameters, 104B activated
- 1M-token context window
- Terminal-Bench 4.0: 65.7% (Moonshot's figure)
- Claude Opus 5.5: 66.4%; GPT-6 Astra: 58.2% on the same test
- Weights posted to Hugging Face on 4 November 2026
- Licence: name display above 100M monthly users or $20M monthly revenue