Moonshot AI

Moonshot AI is the Beijing lab behind the Kimi family. Kimi K3, released July 16, 2026, is the largest open-weight system anyone has shipped: 2.8 trillion total parameters across 896 experts with roughly 16 active per token, a 1 million token context window, and native vision. API pricing is $0.30 per million cache-hit input tokens, $3.00 on a cache miss, and $15.00 output, which works out to the cheapest normalized cost per task among the index leaders. Moonshot reports roughly 57 on the Artificial Analysis Intelligence Index, first place on Frontend Code Arena at 1,679 points ahead of Claude Fable 5, and 93.5 percent on GPQA Diamond, the strongest open-weight result on record; all vendor-reported. Moonshot is also at the center of the July 2026 distillation dispute, after OSTP director Michael Kratsios accused it of distilling Anthropic's Fable model and training on export-controlled Nvidia GB300 servers reached through Thailand. Independent experts have questioned the distillation half of that claim. K3 ships natively in MXFP4 with no FP16 checkpoint, and even at 4-bit it needs roughly 1,450GB of VRAM, so downloadable does not mean runnable without a rack.