Kimi K3.1 Pro open weights: Moonshot's 2.8T model lands on Hugging Face
Moonshot published the open weights for Kimi K3.1 Pro on November 4 as promised at launch: a natively multimodal agentic mixture-of-experts with 2.8 trillion total parameters, 104 billion active per token, 896 experts, a 1M-token context window, shipped in MXFP4 at roughly 1.56TB.
Why it matters
The largest real open-weight release on the Hub, ahead of Qwen3.8-2.4T and DeepSeek-V4-Pro; Gated DSA sparse attention in 24 of 93 layers speeds long-context work, and cyber-offensive content was removed from training data before release.
What you could build with it
A startup building long-document review tools could self-host Kimi K3.1 Pro behind a retrieval pipeline to analyse thousand-page contract sets in one pass, returning clause-level answers with citations instead of chunking documents across many calls.
Does it hold up?
Usable for well-resourced teams today via the hosted API and vLLM/SGLang serving recipes, but self-hosting needs roughly 1.56TB of weights and multi-GPU serving, so most builders will start on the hosted API.
Built with Kimi K3.1 Pro open weights
- Moonshot AI releases 2.8T-parameter Kimi K3.1 Pro weightsworldattention · Daily briefing noting the 2.8T total / 104B active size, 1M context, MXFP4 shipping at about 1.56TB, and a 65.7% Terminal-Bench 4.0 score in Moonshot's own tests.
- Moonshot releases the open weights of Kimi K3.1 Prolaura-martel · Launch writeup covering the custom Kimi K3.1 Pro License, MXFP4 weights with MXFP8 activations, and the October 23 launch that topped 6 of 14 benchmarks.
- Kimi-K3.1-Pro deploy log entrydeployedbyai · Deployment log tracking the Hugging Face publication, noting the model reached its index two days after Moonshot published it.
- vllm-project/vllmgithub · Kimi K3.1 Pro ships with published vLLM deployment recipes for multi-GPU serving.
- sgl-project/sglanggithub · SGLang deployment recipes for Kimi K3.1 Pro were published alongside the weights.
Learn more
First spotted on hf: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72NVIDIA Cosmos 3.5 world models: now openly availableNVIDIA's Cosmos 3.5 family of open world foundation models for robots, autonomous vehicles and vision agents is now…models · JEV 0.71
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.