Moonshot AI open-sources Kimi K3.1 Pro: 2.8T-param MoE with 1M-token context
Kimi_Moonshot · x · 2026-11-05
Moonshot AI has released the open weights of Kimi K3.1 Pro on Hugging Face.
Key facts:
- A natively multimodal agentic model post-trained from Kimi K3 via large-scale joint RL across hundreds of downstream domains
- MoE architecture with 2.8T total params, 104B activated per token, 93 layers, and a 1,048,576-token context window
- Uses KDA + Gated DeepSeek Sparse Attention (replacing K3's full-attention Gated MLA layers) for efficient long-context modeling; MXFP4/MXFP8 quantization-aware training
- Moonshot claims substantial gains in coding, 3D creation, financial research, and factual accuracy
The company says it will keep advancing open intelligence.
Related event: Moonshot AI Open-Sources Kimi K3.1 Pro: 2.8T-Parameter MoE with 1M Context(19 posts)→
More from Models
- Together AI hosts live walkthrough of Kimi K3.1 Pro with Moonshot AI — togethercompute · 2026-11-05
- Model selection is a routing problem: match model size to task ambiguity, says Together — togethercompute · 2026-11-05
- Mistral Large 4 1T hits HF Inference Endpoints at $10/hr on 4x A100 — ngxson · 2026-11-05
- Small model 1017 loses evals to Haiku but "sometimes has frontier-level ideas," argues researcher — teortaxesTex · 2026-11-05
- vLLM ships Day-0 support for Kimi K3.1: 61.2% on long-horizon coding bench, RL-ready stack — vllm_project · 2026-11-05
- Tencent's Hy4 benchmarks close in on Claude Opus 5.5 and GPT 6.1 Sol, topping several leaderboards — heypearlai · 2026-11-05