Moonshot AI has released the weights and technical report for Kimi K3 on Hugging Face and GitHub. The company is also open-sourcing parts of its infrastructure, including high-performance attention kernels, an MoE communication library, and tools for running AI agents at scale. Moonshot claims this architecture delivers 2.5 times more intelligence per unit of compute. Since its announcement in mid-July 2026, the model has scored close to Western frontier models such as Fable 5 and GPT-5.6 Sol on popular benchmarks, but at a slightly lower cost.
However, an independent test by the UK’s Cyber Institute found that Kimi K3’s cyber capabilities lag far behind those of frontier models. The same is true of its math skills. Both gaps could suggest that the model relies on distillation, a technique where a smaller model learns from the outputs of a more capable one. Chinese models often face this accusation, while American open-weight advocates increasingly view distillation as legitimate. This move adds another layer to the global competition over model efficiency and transparency.
* Technical report available on GitHub
* Infrastructure tools include MoE communication library
* Independent tests show gaps in cyber and math skills
