Moonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M Context

Moonshot AI released Kimi K3, a 2.8-trillion-parameter open Mixture-of-Experts model with 1-million-token context and native vision. It introduces Kimi Delta Attention and Attention Residuals, achieving about 2.5x better scaling efficiency than its predecessor K2 while leading in several benchmarks. Accessible via API at $0.30–$15 per million tokens, K3 supports OpenAI SDK integration and targets long-horizon coding and reasoning tasks.