Documentation v1.28.1 for the Kimi K3 weights release catalogs 518 verified claims spanning model architecture, training methodology, benchmark performance, and deployment considerations. The GPQA-Diamond gap of 0.60 (K3 93.54 vs frontier 94.14) positions K3 as near-frontier in graduate-level reasoning. The permissive Kimi K3 License and 96 safetensors shards on HuggingFace enable broad research access.