Moonshot launches Kimi K3, a 2.8T parameter open-weights model

Moonshot AI has launched Kimi K3, a frontier-class open-weights model with 2.8 trillion parameters and native multimodal input. The model features Kimi Delta Attention (KDA) for faster decoding in long contexts and is positioned for long-horizon agentic coding workflows.

• Kimi K3 uses LatentMoE with 16 activated experts out of 896, achieving up to 6.3x faster decoding in million-token contexts.
• It ranked #1 in Frontend Code Arena with a 76% pairwise win rate and scored 57 on the AA Intelligence Index.
• Artificial Analysis reported a 21% reduction in output tokens compared to K2.6, with an average cost of $0.94 per task.
• Open weights are promised by July 27, 2026, with vLLM adding day-0 support for KDA prefix caching.

The launch is viewed as a significant open-model milestone, offering competitive performance against top closed models while providing cost and efficiency advantages.