Kimi K3: 2.8T Parameter Model Ships With 1M Context
Kimi K3 launches at 2.8 trillion parameters, 1 million context and native multimodal support, delivering 6.3x faster decoding via Kimi Delta Attention.
SourceAnalysis
Kimi K3 ships with 2.8 trillion parameters, 1 million context length and native multimodal capabilities, prompting immediate praise from Meta AI lead Soumith Chintala. The model introduces Kimi Delta Attention for up to 6.3x faster decoding at million-token lengths and Attention Residuals that lift training efficiency 25 percent for under 2 percent extra cost. Built explicitly for long-horizon agentic coding and self-evolving workflows, Kimi K3 is live today on Kimi.com, Kimi Work, Kimi Code and the Kimi API, with open weights scheduled for July 27, 2026.
Soumith Chintala
@soumithchintalaCofounded and lead Pytorch at Meta. Also dabble in robotics at NYU.