Meta publishes Muse Glimmer’s technical report: the 30B text-and-image model uses 4-bit quantization for local agents within 32GB of GPU memory, with weights under Apache 2.0.

Figure 2 from Meta’s October 19 Muse Glimmer report, comparing company-reported capability scores with Gemma4-31B and Qwen3.6-27B.

Published 19 Oct 2026 · 23:04 UTC

Meta AI Research ↗ · Source dated 19 Oct 2026

Browse all Live updates · Read Features