Meta Launches 30B Open-Source AI Model for PCs

Meta released Muse Glimmer, a 30-billion-parameter open-source language model optimized to run on personal computers and Macs using a single consumer-grade GPU. Through quantization and other techniques, Meta reduced its memory footprint from 55GB to under 20GB. The model uses speculative decoding for faster responses and was trained to retry failed tasks autonomously. Meta CEO Mark Zuckerberg accompanied the launch with an essay calling for reduced open-source AI regulations and urging labs to grant governments early model access to address cybersecurity risks. Muse Glimmer outperformed comparable models on half of 24 benchmarks tested, covering coding, research, and scientific analysis.