Moonshot AI has released a 24-second teaser video for its upcoming Kimi K3.1 model. The footage shows the model name displayed on an old computer running Kimi Code, signaling an imminent launch, though the company has not yet announced an official release date.
Technical evidence of the new model appeared in September when developers spotted the 'kimi-k3-1' identifier in the Moonshot AI API registry. The platform's preview indicated support for a 1M token context window. Further confirmation emerged on October 19 via a code commit in the SGLang repository, which added support for K3.1 chat encoding and included 107 test cases.
The K3.1 update introduces a new 'medium' level to the existing low, high, and max reasoning intensity settings. According to technical documentation, the model maintains compatibility with K3 tokenizers and special tokens. Analysis of the SGLang repository suggests that a 'moonshotai/Kimi-K3.1' model entity is already present on Hugging Face, though it remains private.
The teaser's focus on Kimi Code suggests that developer workflows remain a priority for the company. Previous iterations of the Kimi model were integrated into GitHub Copilot in August and Amazon AWS Kiro in October. While the K3.1 interface hints at new features like 'make3d' for 3D content and 'OK_COMPUTER_SWARM' for multi-agent collaboration, these have not been confirmed by Moonshot AI.
