Moonshot AI teases Kimi K3.1 with a pre-launch video

Moonshot AI has released a teaser video for Kimi K3.1, the clearest signal yet that a new version of its Kimi model line is close to launch. Chinese outlets 36kr and Sohu both reported the video on October 23, framing their stories around the same idea: K3.1 is coming.

The reports carry very little beyond that. What they establish is that a teaser exists and that the version number K3.1 is now in public use. Model size, context length, reasoning performance, whether weights will be open, and API pricing do not appear in either account.

Teaser videos have become a standard ritual for Chinese model developers. Compared with a technical blog post, a short clip travels further on social platforms and shifts attention toward when the model can be used rather than what its benchmark table looks like.

K3.1 is a decimal iteration on the K series rather than a new generation. Updates of this kind usually do one of two jobs: fixing problems that surfaced in real use, or closing a capability gap a competitor has opened. Which one this is cannot be determined from what has been published.

For developers and enterprise buyers, the question that matters is whether K3.1 changes anything at the interface level, from context limits and tool-calling behaviour to latency and price per token. A teaser answers none of that, and a launch event may not either.

Expect more clues in the coming days, most likely through further short videos or countdown posts, with verifiable numbers arriving only at the formal launch. Until then, the teaser is a reliable signal of timing and a weak signal of capability.