Moonshot AI has released the full 2.8-trillion-parameter Kimi K3 model on Hugging Face under an open-weight license. The model uses a Mixture-of-Experts architecture, activating only 104 billion parameters per request, and is compressed to roughly 1.56TB using MXFP4 weights. Benchmark-wise, K3 competes with closed-source frontier models like Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol, while being significantly cheaper on API pricing. Self-hosting remains impractical for most — a community project called WASTE managed to run it on a 64GB MacBook Pro, but throughput was only 0.32–0.34 tokens per second due to SSD streaming bottlenecks. The release is seen as a milestone for the open-weight AI ecosystem, signaling that the gap between open and closed models is narrowing.

4m read timeFrom xda-developers.com
Post cover image
Table of contents
Almost all of Kimi K3 is now on Hugging FaceKimi K3 can go against any model out thereSelf-hosting it is a distant dream, though
215 Impressions1 Comment