Moonshot AI launches Kimi K3, open-weight frontier model with 2.8T params

Kimi K3 features 2.8 trillion parameters, 1 million token context, and native multimodal capabilities. It uses Kimi Delta Attention for up to 6.3x faster decoding and Attention Residuals for ~25% improved training efficiency. The model is open weight and available via partnerships with Together AI, Nebius, and as a GGUF quant.