Moonshot releases Kimi K3.1 Pro: 2.8T parameters, open weights

Moonshot AI published Kimi K3.1 Pro on 2 November 2026: a 2.8T mixture-of-experts model for long coding and office tasks, with a 1M-token context and 4-bit weights of about 1.56 TB.

  • New release
  • 2 min read
  • Published
  • Updated

Moonshot AI published the weights of Kimi K3.1 Pro on 2 November 2026. It is the first model in the K3.1 series and the follow-up to Kimi K3, which came out on 27 July. Moonshot calls it its most capable open-weight model of the 3-trillion-parameter class.

What is new

  • Post-training. K3.1 Pro is post-trained from K3 with large-scale reinforcement learning across hundreds of downstream domains.
  • Size. It is a mixture-of-experts model with about 2.8 trillion parameters in total and roughly 104 billion active for each token: 16 of 896 routed experts, plus 2 shared experts.
  • Attention. Most layers use Kimi Delta Attention, a linear-attention design. Compared with K3, Moonshot replaced the full-attention layers with Gated DeepSeek Sparse Attention, which makes very long inputs cheaper to process.
  • Inputs. It accepts text and images, writes text, and has a context window of 1,048,576 tokens.
  • Weights. They are distributed in MXFP4, a 4-bit format that Moonshot says the model was trained with from the start, so the 1.56 TB download is already the compact version.

Moonshot reports gains over K3 in coding, 3D creation, financial research and factual accuracy. Those numbers come from Moonshot’s own model card and test harnesses and have not been reproduced independently.

Why it matters

Kimi K3.1 Pro is one of the largest sets of weights anyone can download, and Moonshot built it for long coding and office tasks. It is not a local model: by our estimate it needs about 1.7 TB of memory, which puts it in the multi-node cluster tier, so most people will reach it through Moonshot’s API. Teams that already run large GPU clusters can host it themselves. For another frontier-scale option, see our comparison with Qwen3.8-Max.

License

K3.1 Pro uses the Kimi K3 License terms. It is MIT-style: most companies can use it commercially without asking. A business that sells hosted access to the model needs a separate agreement once its combined revenue passes US$20 million over any 12 months, and a very large commercial product, with over 100 million monthly active users or over US$20 million of monthly revenue, must show “Kimi K3” prominently in its interface.

Sources

Models in this story

Model Released Parameters Context License Smallest variant needs
Kimi K3.1 Pro Moonshot AI 2.8T 104B active 1M Kimi K3 License (Conditional) 1.72 TB Multi-node cluster
Kimi K3 Moonshot AI 2.8T 104B active 1M Kimi K3 License (Conditional) 1.72 TB Multi-node cluster