BEIJING, November 5, 2026 — Moonshot AI has released the weights of its Kimi K3.1 Pro model on Hugging Face, making its flagship multimodal model available for independent deployment. This move is significant for the AI industry, as it provides a free and open alternative to expensive closed APIs.
Model Details
The Kimi K3.1 Pro model uses a mixture-of-experts architecture with 2.8 trillion total parameters and approximately 104 billion active parameters per token. It supports a context window of up to 1 million tokens, making it suitable for tasks that require long-range dependencies. The model has been post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains, which has improved its performance in coding, 3D creation, financial research, and other agentic capabilities.
Benchmark Performance
Kimi K3.1 Pro has been evaluated on several benchmarks, including Terminal-Bench 4.0, SWE Marathon v1.1, and BrowseComp. According to the results, the model has achieved scores that surpass or match those of leading models like GPT-6 Astra and Claude Opus 5.5. For example, in Terminal-Bench 4.0, Kimi K3.1 Pro recorded a score of 65.7%, outperforming GPT-6 Astra’s 58.2% and slightly behind Claude Opus 5.5’s 66.4%. In SWE Marathon v1.1, the model achieved a score of 51.7%, clearly outperforming GPT-6 Astra’s 42.5% and Claude Opus 5.5’s 46.3%.
Licensing Terms
The Kimi K3.1 Pro model is released under the Kimi K3.1-Pro License, which permits free use, modification, redistribution, and commercial use. However, the license requires companies that offer the model as a hosted API service with more than $20 million in trailing 12-month revenue to sign a separate commercial agreement with Moonshot AI. This condition is aimed at preventing large-scale API providers and giant platforms from taking the model without contributing to its development.
Hardware Requirements
Running the Kimi K3.1 Pro model at usable speed requires significant hardware resources, including approximately eight Nvidia B300 GPUs and 2.3 terabytes of combined GPU memory. This makes it accessible to hyperscalers, well-funded labs, and inference providers who want to cut out the API markup.
Impact
The release of Kimi K3.1 Pro has significant implications for the AI industry, as it provides a free and open alternative to expensive closed APIs. According to Lucky AI Lab, the model’s performance is comparable to or better than that of leading models like GPT-6 Astra and Claude Opus 5.5, but at approximately 20% of the price. This changes the paradigm of technology selection, where high performance was previously synonymous with using high-cost closed APIs.
The model’s capabilities and licensing terms make it an attractive option for companies that want to incorporate AI into their services without incurring significant costs. As noted by BigGo Finance, the release of Kimi K3.1 Pro has entered a phase of competing to address the misuse risks that come with high capability, with Chinese players releasing the weights of high-performance models for free while prioritizing safety assurance. On November 4, Huawei announced inference support for Kimi K3.1 Pro via vLLM Ascend and SGLang on Atlas A3 and Ascend 950 systems.
Sources
- moonshotai/Kimi-K3.1-Pro · Hugging Face
- Kimi (AI)
- Moonshot AI releases Kimi K3.1 Pro model weights · TechNode
- Moonshot AI Releases Full Weights For 2.8 Trillion Parameter Kimi K3.1 Pro – Startup Fortune
- Moonshot AI Releases Full 1.56TB Weights of Kimi K3.1 Pro for Free, Permits Commercial Use With Conditions — BigGo Finance
- 3T Parameter Model Released at 20% of the Price: The Cost Advantage of Open Weights｜Lucky AI Lab
Photo: Julian Herzog (Website) (CC BY 4.0)
