Kimi ships K3.1 Pro and Highspeed, with full weights planned by November 4

Kimi has released K3.1 Pro, a post-trained upgrade to K3 with a one-million-token context window. It also released K3.1 Pro Highspeed, which the company says delivers two to three times K3.1 Pro's speed for more responsive interactions.

Both versions are available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. That gives teams already using the Kimi API or Kimi Code a concrete new option to benchmark on real repository fixes, research, and agent tasks. The published benchmark results are vendor claims and should not replace an evaluation on a team's own tasks, tool calls, latency, and recovery conditions.

Although the announcement frames the release as part of an open-intelligence path, full model weights are planned by November 4. Teams that need local or controlled-environment deployment should not treat the API launch as self-hosting availability, or infer hardware and commercial-use terms before the weight package and license are available.

Start with a bounded, reversible evaluation of quality, speed, and cost per completed task before moving a production workflow. If deployment control is a hard requirement, wait for the complete weights and license text.

Related catalog entry