Kimi-K3 Releases on HuggingFace 6/27
What to expect
- New architecture built on Kimi Delta Attention and Attention Residuals
- Native agentic capabilities: tool calling, browsing, and multi-step planning
- Extended context window designed for repository-scale code understanding
- Open weights, released right here on this page
Good lord that’s a lot of VRAM. Hopefully soon there’s another phase of finding good ways to distill this recent progress into smaller models I can actually run locally.