Reportedly: Moonshot AI releases Kimi K3.1 Pro, a 3-trillion-parameter model with 1M context, ranking 4th on Artificial Analysis
Model: Kimi K3.1 Pro, Kimi K3 · availability, license and releases
Reported by Zhidongxi · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Moonshot AI released Kimi K3.1 Pro, a 3-trillion-parameter model with a 1M-token context window. It ranks 4th overall and 1st among Chinese models on Artificial Analysis, scoring 11 points above Kimi K3. Input price is 13 yuan (approx. $2) per million tokens, output 68 yuan (approx. $10), 35% and 32% lower than K3 respectively, Zhidongxi reported.
China context
- Original name
- 月之暗面
- Outside China
- Not stated in the sources yet
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can evaluate the model via API now; open weights are promised by November 4, 2026, which would allow local deployment if the license permits.
- For investors
- Moonshot AI's aggressive pricing at 20% of GPT-6 Astra's cost signals a strategy to capture global API market share, potentially pressuring margins for Western competitors.
Translated from Chinese. Quotes and facts link to the original sources.
Moonshot AI launched Kimi K3.1 Pro, its latest flagship model, with 3 trillion parameters and a 1-million-token context window. The model is post-trained from Kimi K3 and will be open-sourced by November 4, 2026. On the Artificial Analysis leaderboard, it ranks 4th overall and 1st among Chinese models, surpassing Claude Fable 5.1 and GPT-6 Astra. The company highlights improvements in programming, 3D creation, and financial research. Demonstrations include building game scenes in Unreal Engine, creating a working turbofan engine model in FreeCAD, and replicating Singapore Airport's indoor rainforest in Blender. The model is available via Kimi's website, Kimi Work, Kimi Code, and API. A high-speed version offers 2-3x faster inference at double the price. Pricing is 13 yuan (approx. $2) per million input tokens and 68 yuan (approx. $10) per million output tokens, which is 20% of GPT-6 Astra's list price and 50% of Claude Opus 5.5's.
Kimi K3.1 Pro is a 3-trillion-parameter model with a 1M-token context window, post-trained from Kimi K3. It demonstrates capabilities in 3D modeling via tool integration (Unreal Engine, Blender, FreeCAD) and scores 11 points higher than K3 on Artificial Analysis. The high-speed version achieves 2-3x faster inference.
Moonshot AI's price cut to 20% of GPT-6 Astra's cost directly pressures OpenAI and Anthropic on API pricing, forcing competitors to justify premium rates or lower prices for developers.
For enterprises, the 1M-token context and lower price (35% input, 32% output reduction vs K3) reduce cost for long-context tasks like code analysis and document processing. The high-speed version offers a premium tier for latency-sensitive applications.
Open-sourcing by November 4, 2026 will be a key signal: if weights are released under a permissive license, adoption among developers outside China could accelerate. Independent verification of the Artificial Analysis score and 3D modeling claims is needed.