Moonshot AI open-sources Kimi K3 weights and its training infrastructure
Moonshot AI released weights and the technical report for Kimi K3, a 2.8-trillion-parameter model, plus attention kernels and an MoE communication library, claiming 2.5x more intelligence per unit of compute than its predecessor. A model benchmarking near Fable 5 and GPT-5.6 Sol is now downloadable, though the UK Cyber Institute found its cyber and math skills trail frontier models — a gap consistent with distillation, which US open-weight advocates increasingly treat as legitimate.
Source: the-decoder.com ↗
Moonshot AI claims the new architecture delivers 2.5 times more intelligence per unit of compute.
Moonshot AI
Why this matters
- → Open-weight frontier model now downloadable, reshapes what developers can build locally
- → Distilled model shows open-weight strategies gaining legitimacy vs. training from scratch
Open-weight frontier arrives