Moonshot AI releases Kimi K3: 2.8T parameter open frontier model

The 2.8T-parameter model features 1M context, native multimodal, Kimi Delta Attention (6.3x faster decoding), and Attention Residuals (~25% training efficiency). It ranks #1 on SpreadsheetBench 2, scores 57 on AI Intelligence Index comparable to Opus 4.8 and GPT-5, and tops Fable 5 on several coding benchmarks. Open-weight release positions it as a strong competitor to leading frontier models.

Featured · Yang Zhilin