Kimi doesn’t “go all-in” on video generation; it focuses on raising the ceiling of intelligence—this is the real direction for large models.

Chainnews message: Kimi’s co-founder and head of the algorithm team, Zhou Xinyu, said that Kimi currently has no plans to develop a video generation model. At this stage, the team is more focused on improving the model’s intelligent upper limit. He believes that while video generation is useful, it does not meaningfully help improve a model’s intelligence, whether from a theoretical or practical standpoint. Kimi K3 has just been released, and the key capabilities listed by the official are still software engineering, knowledge work, deep reasoning, and image understanding. Kimi’s multimodal roadmap currently leans more toward understanding and reasoning rather than generating videos.