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

Kimi K3 features 2.8 trillion parameters, 1 million token context, and native multimodality. It uses Kimi Delta Attention for up to 6.3x faster decoding and Attention Residuals for ~25% higher training efficiency. Early benchmarks rank it 3rd on VoxelBench and show it competitive with Opus 4.8 and GPT-5.