Kimi K3 custom GGUF quant works at Q3_K_S, 1.1 TB on disk

Community devs built dynamic GGUF quants of Kimi K3 from original weights using a llama.cpp fork; Q3_K_S runs at 1114.76 GiB on disk, with Q1 and Q2 in progress.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Cable Management extension for ComfyUI published
- Reddit user shares Star Trek-style clip made with Minimax H3
- a16z podcast: AI models now exploit vulnerabilities, not just find them
- Claude Opus 5 praised by student, then fails simple chart task
- Airbnb tests AI-powered search with user-controlled toggle