Chinese lab Moonshot AI has released Kimi K3, an open-weights large model that narrows the performance gap with leading U.S. systems. K3 supports a 1 million-token context window and posts strong benchmark results in web research, spreadsheets, frontend design, and long-form coding, bringing attention to non-U.S. labs’ rapid progress.
On public intelligence rankings, K3 scores near the top — sitting just behind Anthropic’s Fable 5 and OpenAI’s GPT-5.6 Sol—while improving markedly over Moonshot’s previous K2.6. In demonstrations, the model ran a 48-hour solo workflow to design and verify a tiny chip, showing practical, sustained reasoning at scale.
Pricing and openness set K3 apart: Moonshot prices it competitively with leading commercial models and plans to publish the model weights soon. That combination of performance, price, and imminent open weights could broaden access and speed adoption by researchers and developers worldwide.
The release renews questions about how far behind China—or any open-source lab—really is from U.S. frontier teams. One model won’t settle that debate, but K3 makes it clear the global AI landscape is now multi‑polar and intensely competitive.
GLOBAL AI IS NO LONGER A TWO‑HERO STORY — THE RACE IS OPEN.
Sanjay Sahay
Have a nice evening.