Moonshot AI published the full weights of its Kimi K3.1 Pro model on Hugging Face on November 4, meeting the deadline it set when it launched the model in October and handing one of the largest open-weight AI models ever released to the public.
The Chinese AI company announced the release on X at 16:00 UTC, writing simply: "Releasing the model weights of Kimi K3.1 Pro." The repository, moonshotai/Kimi-K3.1-Pro, holds 96 weight files totalling about 1.56 TB, with no access gate. Moonshot had launched Kimi K3.1 Pro on October 23, calling it "the first model in our new K3.1 series and our most capable 3T-class open-weight model to date."
The model card lists a natively multimodal, mixture-of-experts design with 2.8 trillion total parameters, 104 billion of them active for each token, selected from 896 experts, and a context window of 1,048,576 tokens. The weights are stored at 4-bit precision after quantization-aware training. Moonshot recommends the vLLM, SGLang and TokenSpeed inference engines, and the simplest setup requires eight data-center GPUs, so most users are expected to keep using the hosted version, priced at $2 per million input tokens and $10 per million output tokens.
In its own benchmark table, Moonshot says Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, a test of coding agents on real terminal tasks, against 66.4% for Anthropic's Claude Opus 5.5 and 58.2% for OpenAI's GPT-6 Astra. Across the 37 non-internal benchmark rows where both models have a score, Kimi K3.1 Pro comes out ahead of Claude Opus 5.5 on 16 of them, though Opus 5.5 leads by more than 20 points on three rows.
The weights ship under the Kimi K3.1-Pro License, Moonshot's own license rather than a standard open-source one. It allows most users to use, copy, change, fine-tune and sell the model, but a company running a model-as-a-service business with more than $20 million in revenue over any 12 months must sign a separate agreement with Moonshot before using it commercially.
Moonshot said it traced rising exploit skill in the model to cyber content in its general mid-training corpus and reinforcement-learning environments, and removed both before opening the weights. The release makes Kimi K3.1 Pro one of the largest real open-weight model releases on Hugging Face, ahead of other recent large models including DeepSeek-V4-Pro.
