Moonshot AI releases Kimi K3.1 Pro weights, a 2.8T-parameter open model with a million-token context
Kimi K3.1 Pro model weights are out
Kimi K3.1 Pro is Moonshot AI's most capable open-weight model yet: a 2.8T-parameter mixture-of-experts with 104B activated parameters, native multimodality, and a 1,048,576-token context window. Post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of domains, it swaps full-attention layers for Gated DeepSeek Sparse Attention and posts strong agentic results, including 70.0 on HLE-Full with tools and 92.1 on BrowseComp.
Compared with Kimi K3, we replace the full-attention Gated MLA layers with Gated DeepSeek Sparse Attention (Gated DSA) layers for more efficient long-context modeling.