Overview
- Moonshot announced Kimi K3 on July 17, describing it as a 2.8‑trillion‑parameter open‑weight model with a 1‑million‑token context window designed for long‑horizon reasoning and programming tasks.
- Independent evaluators reported strong task‑level results that in some benchmarks place Kimi K3 near Anthropic’s Fable 5 while confirming it still falls short of the very top proprietary systems on overall user experience.
- Anthropic has publicly accused Moonshot of large‑scale illicit extraction and distillation of its model outputs, a claim that adds potential legal and IP risks to the launch.
- Markets moved sharply after the announcement, with shares of several Chinese AI firms plunging on the Hong Kong exchange, and Moonshot says it will publish full model weights for download by July 27.
- Experts note that running a 2.8‑trillion‑parameter model requires costly specialized hardware, Moonshot’s recent US$2 billion fundraising backs rapid scaling, and the release could speed adoption of Chinese open models while prompting tighter regulation and export controls.