Moonshot AI has released Kimi K3, a massive 2.8-trillion parameter open-weight model that is topping key AI benchmarks.
The model scored 2,840 on Towards AI's Writing Elo benchmark, surpassing Claude Fable 5's 2,760. It also claimed the top spot on Arena AI's Frontend Code Leaderboard.
On the Artificial Analysis Intelligence Index, which aggregates nine independent evaluations, K3 scored 57. That places it third, just behind Claude Fable 5 at 60 and GPT-5.6 Sol at 59.
K3 uses a mixture-of-experts architecture with 896 expert subnetworks. It includes a one-million-token context window and native image and video understanding.
The model is priced at $3 per million input tokens and $15 per million output tokens-the same as Anthropic's mid-tier Claude Sonnet 5. This represents near-frontier performance at a mid-tier price point, a significant cost advantage for API developers.
Moonshot AI president Yutong Zhang attributed the success to architectural innovation under hardware constraints, stating the company focused on fundamental research and efficiency.
However, K3's hallucination rate increased to 51% on the AA-Omniscience benchmark, up from 39% in its predecessor. The model can also be "excessively proactive" during autonomous tasks.
Full model weights will be released on July 27 under a modified MIT license, making K3 the largest freely available AI model in history.