San Francisco — Open-weight AI models reached 75% of frontier accuracy on a 64GB laptop. Quantization compresses models to fit consumer RAM, enabling offline coding loops without per-token billing or
Hacker News