Overview
- Moonshot unveiled Kimi K3 as a 2.8 trillion‑parameter open‑weight model on July 16, positioning it as one of the largest publicly released systems and claiming performance near top U.S. models.
- Unprecedented user demand quickly saturated Moonshot’s GPU clusters and the company said over a 48‑hour period it would pause new consumer subscriptions to protect existing paid users and add capacity.
- Moonshot still plans to publish Kimi K3’s full model weights on July 27, which would let developers download, run and modify the model themselves and could shift heavy inference loads off Moonshot’s servers.
- Independent and vendor benchmarks report strong results for Kimi K3 on coding, long‑context and some vision/agent tasks, but verification is limited and performance varies by task and test.
- The launch moved markets and strategy: Alibaba and Tencent shares rose while chip stocks fell, Moonshot is seeking fresh capital and a possible Hong Kong IPO, and policymakers are revisiting export controls, distillation risks and how infrastructure limits diffuse capabilities.