The repository is moonshotai/Kimi-K3.1-Pro. It is 96 safetensors shards, about 1.56 TB, in MXFP4 weights with MXFP8 activations. The model card says it was trained that way, so this is the precision Moonshot serves. The card recommends vLLM, SGLang or TokenSpeed to run it. It is post-trained from Kimi K3, and the one architecture change is that the full-attention layers became Gated DeepSeek Sparse Attention layers, for cheaper long context.
The license is the Kimi K3 license with a new name, according to ArtificialWatch, which compared the two files. Use, modification, fine-tuning and redistribution are free. A company selling model access as a service with more than $20 million in revenue over any 12 months needs a separate agreement with Moonshot. A product with more than 100 million monthly users or more than $20 million in monthly revenue has to display "Kimi K3.1-Pro" prominently.
ArtificialWatch ranks it the highest model with downloadable weights on its Bench Index (69.9, #11 of 44). Moonshot hasn't said whether the faster Highspeed variant is a separate checkpoint, and no Highspeed repository is posted.