Chinese lab Moonshot AI has just released Kimi K3, an open-weights model that pulls within a few benchmark points of frontier systems such as Fable 5 and GPT-5.6 Sol, and even beats both on certain tasks, all at a fraction of frontier prices.
The model ships with a 1 million token context window. In one internal demo, K3 worked solo for 48 hours to design and verify a tiny chip that runs a mini version of itself, hitting 8,700 tokens a second in simulation.
Kimi K3 is priced at $3 per million input tokens and $15 per million output tokens, matching Claude 5 Sonnet. Moonshot AI says the model’s weights will be published by July 27.
The release raises a sharp question: did Moonshot just score 2026’s DeepSeek moment? Pulling right up on frontier models is no small feat, especially for a soon-to-be open-weights model at this price point. Dario Amodei’s “6-12 months behind” estimation for both China and open source suddenly looks more like just a single release cycle.
For users watching the AI race, the broader signal is clear: the gap between top proprietary models and well-resourced open-weight labs is shrinking faster than expected. Expect benchmark tables and cost comparisons to shift in the coming weeks.
