Chinese AI lab Moonshot AI has announced Kimi K3, describing it as their most capable model to date with 2.8 trillion parameters. The model is currently available via the website and API, with an open weight release promised by July 27, 2026.

  • Kimi K3 reportedly beats Claude Opus 4.8 max and GPT-5.5 high in self-reported benchmarks, while losing to Claude Fable 5 and GPT-5.6 Sol.
  • Artificial Analysis reports an Elo of 1547 on long-horizon knowledge work, behind only Claude Fable 5.
  • Pricing is set at $3/million input tokens and $15/million output tokens, making it the most expensive model from a Chinese lab to date.
  • The model leads Arena.ai's Frontend Code arena and uses 21% fewer output tokens than its predecessor K2.6.

The article notes that Kimi K3's high cost is driven by significant reasoning token usage, which can make simple tasks expensive compared to other models.