Moonshot Publishes Kimi K3 at 2.8T Parameters — First Open Model in the 3-Trillion Class, MXFP4 From Day One

HuggingFace·high signal

Moonshot AI publicly released Kimi K3 on July 16, a 2.8T-parameter MoE routing 896 experts with 16 active per token, native multimodal input, and a 1M-token context — with full open weights promised by July 27. It ships in MXFP4 (4-bit float with per-block scaling), which puts the full weight footprint around 1.4 TB rather than the ~5.6 TB bf16 would demand. API pricing is $3.00/M input on cache miss ($0.30 on hit) and $15.00/M output, meaning the quantization choice is what makes self-hosting even theoretically discussable at this scale.