THE ESSENTIALS IN 30 SECONDS

  • Moonshot AI released the Kimi K3.1 Pro weights on Hugging Face on November 4, keeping the date it set when the model reached its API in October.
  • The model card describes a mixture-of-experts design with 2.8 trillion total and 104 billion active parameters, a 1,048,576-token context and image input.
  • The Kimi K3.1-Pro License is permissive, but large model-as-a-service operators need a separate agreement and very large products must display the model's name.

Moonshot AI, the company behind the Kimi assistant, released the weights of Kimi K3.1 Pro on Hugging Face on November 4, RuntimeWire reports. The model had been available only through Moonshot's API and apps since late October, when the company said the weights would follow on this date. Developers can now download and host the model themselves, subject to a licence written by Moonshot.

What the model card describes

According to the Hugging Face page, Kimi K3.1 Pro is a mixture-of-experts model with 2.8 trillion total parameters, of which 104 billion are active per token. It routes each token to 16 of 896 experts, supports a context of 1,048,576 tokens and accepts images through a 401-million-parameter vision encoder. The weights were trained with quantization in mind and are shipped in MXFP4 format. Moonshot recommends vLLM, SGLang or TokenSpeed for inference. VentureBeat reported in October that Moonshot attributes the gains over Kimi K3 mainly to additional post-training, including reinforcement learning across many task domains, rather than a new architecture.

Benchmarks and pricing

The model card compares Kimi K3.1 Pro with GPT-6 Astra, Claude Opus 5.5 and other systems. These are Moonshot's own figures, and some rely on internal tests. On them, the model leads on BrowseComp (92.1) and on Humanity's Last Exam with tools (70.0), but trails Claude Opus 5.5 on Terminal-Bench 4.0 (65.7 against 66.4). Media coverage of the API launch said Artificial Analysis, an independent benchmarking firm, gave the model a score of 55 on its Intelligence Index. VentureBeat and GenAI News reported API prices of $2 per million input tokens and $10 per million output tokens.

What the licence allows

The Kimi K3.1-Pro License grants broad rights to use, modify, fine-tune, redistribute and sell the software, provided the copyright notice is kept. Two conditions set it apart from standard permissive licences. Companies running a model-as-a-service business with more than $20 million in revenue over any 12-month period must sign a separate agreement with Moonshot before commercial use. Products with more than 100 million monthly active users or $20 million in monthly revenue must display "Kimi K3.1-Pro" prominently in their interface. Internal use is exempt. RuntimeWire notes that these terms limit how open the release really is, and independent evaluations of the downloadable weights have yet to appear.