Moonshot Releases the Model Weights of Kimi K3.1 Pro
Releasing the model weights of Kimi K3.1 Pro: t.co/tJXcYUeSCj We'll keep advancing open intelligence.
Context
Open weights for a model that sits next to frontier labs on benchmarks let anyone run, study and fine tune a top model on their own hardware.
Kimi is the model family from Moonshot AI, a Chinese lab. Kimi K3.1 Pro already appears among the frontier models compared in today's Tencent Hy4 charts. Open weights mean the trained model files can be downloaded and run outside Moonshot's own service.
Deep dive
Kimi.ai (@Kimi_Moonshot), the official account of Moonshot AI's Kimi models, posted on Nov 4 that it is releasing the model weights of Kimi K3.1 Pro, with a link to the files. The company added that it will keep advancing open intelligence.
The post is short and does not list size, license or benchmark scores in its text. Those details sit behind the linked page and should be checked there before relying on them.
The release still stands out. Kimi K3.1 Pro is already treated as a peer of frontier models: Tencent's Hy4 launch earlier today compared itself against it alongside Claude Opus 5.5 and GPT 6.1 Sol. Putting the weights of a model at that level in public means researchers, companies and hobbyists can run it on their own machines, fine tune it, and inspect how it behaves, without sending data to Moonshot.
The image with the post shows a collage of nature photos and a retro terminal running Kimi Code with Kimi K3.1 selected as the model.
The post passed 390,000 views and 6,000 likes within hours, more than any other lab release in today's scan.
What to expect next
Watch for independent tests of the open weights against the hosted version, and for the license terms. Inference providers will likely add the model within days, and quantized versions for smaller machines usually follow soon after. US open model makers, and Tencent with its promised Hy4 weights, will be compared against it, and policy debates about Chinese open models may pick it up.