Overview
- Meta released Muse Glimmer on Monday, Aug. 10, 2026, publishing 30‑billion‑parameter weights under the Apache 2.0 license so third parties can run and modify the model freely.
- Muse Glimmer is distributed with 4‑bit quantization that Meta says lets the model run on a single 24GB or 32GB GPU with minimal performance loss, enabling fully local agent and conversational workflows.
- Meta reported tests on MacBook M4 Max, M5 Max and RTX 5090 hardware and said Muse Glimmer matches other 30B‑class models on mainstream benchmarks compared with models such as Gemma4‑31B and Qwen3.6‑27B.
- In a concurrently published essay, CEO Mark Zuckerberg urged U.S. policy changes to ease data and model‑use restrictions, pledged to publish weights for Muse Spark 1.2, and announced a $1 billion community fund.
- The move reduces dependence on cloud APIs, could improve user privacy and lower costs for developers who run models locally, and raises competitive pressure from already strong Chinese open‑weight models such as Kimi K3 and Qwen3.8‑Max.