Kimi K3.1
A next-generation open-source model from Moonshot AI currently in pre-release preparation. It is designed to address known limitations of K3, with the goal of matching or surpassing top-tier models in the Claude and GPT series while reducing token pricing.
Timeline (last 90 days)
Moonshot AI may release a new model in mid-October, having previously teased Kimi K3.1.
Developers discovered an unreleased Kimi K3.1 model identifier in the model registry of Moonshot AI's API platform and were able to successfully invoke it.
The visible information for Kimi K3.1 includes a maximum context length of 1 million tokens, three levels of reasoning intensity (low/high/highest), a Swarm multi-agent collaboration mode, and search and batch processing capabilities.
Moonshot AI's Kimi online configuration has been updated, adding a new K3.1 agent routing model that includes low/high/ultra-high thinking modes, a 1-million-token context window, and support for Swarm architecture.
Kimi K3.1 is approaching release, with online configuration updates showing low, high, and maximum modes and support for a 1 million token context.
Bloggers predict that Kimi K3.1 may surpass Opus 5 and approach the performance level of Fable 5.
Moonshot AI posted a string of mysterious numbers on Chinese social platforms that, when decoded, point to the mathematical constant Pi, interpreted as a hint that Kimi K3.1 is coming soon.
A long string of numbers posted by Kimi's official account, after removing "3.1," was found to match the digits of pi, and has been interpreted as a veiled teaser for Kimi K3.1.
All Facts (3)
Moonshot AI may release a new model in mid-October, having previously teased Kimi K3.1.
50%UnverifiedDevelopers discovered an unreleased Kimi K3.1 model identifier in the model registry of Moonshot AI's API platform and were able to successfully invoke it.
50%UnverifiedThe visible information for Kimi K3.1 includes a maximum context length of 1 million tokens, three levels of reasoning intensity (low/high/highest), a Swarm multi-agent collaboration mode, and search and batch processing capabilities.
50%