A Chinese startup that didn’t exist three years ago just released an AI model that outperforms some of the most powerful proprietary systems on the planet. Moonshot AI’s Kimi K3, an open-weight model packing 2.8 trillion parameters, is now going toe-to-toe with Claude Opus 4.8 max and GPT-5.5 high across multiple benchmarks.
What Kimi K3 actually does
Kimi K3 excels in three areas that matter most for enterprise adoption: long-horizon coding, knowledge work, and reasoning tasks. In independent evaluations, the model scored an Elo rating of 1547 in long-horizon knowledge work. That’s a 732-point jump from the Kimi K2.6 version.
The model only closely trails Claude Fable 5, which sits at the absolute frontier of AI capability right now.
Moonshot AI demonstrated Kimi K3 autonomously editing a teaser video using 56 different source clips, no human in the loop. It handles GPU kernel optimization and frontier physics research.
