Kimi Linear: An Expressive, Efficient Attention Architecture
Kimi Linearattention architecturemachine learningartificial intelligencecomputer sciencehybrid linear attentionfair comparisonsreinforcement learningscaling regimespretrainingdecoding throughputKV cache usage
Author: ronfriedhaber
Date: 7/28/2026
Article Summary:
The paper introduces Kimi Linear, a hybrid linear attention architecture that outperforms full attention in various scenarios, including short-context, long-context, and reinforcement learning (RL) scaling regimes.