K3.1 Pro and the faster K3.1 Pro Highspeed were made available on Kimi.com, Kimi Work, Kimi Code, and the Kimi API; the company said full model weights would be released by November 4.
The release note also describes a separate process for the open-weight version’s cyber capabilities. Kimi says it observed vulnerability-research performance rising as post-training scaled, tracing relevant material to both its mid-training corpus and its reinforcement-learning environment mix. It says it identified and removed cyber-related data from both stages while retaining general-purpose coding tasks.
Kimi says it evaluated vulnerability-research capabilities and applied targeted training controls intended to reduce misuse risk while preserving general usefulness. The post does not provide an independent safety assessment, a complete account of the removed data, the exact controls, or evidence that the released weights cannot be used for cyber misuse.