Meta released Muse Glimmer, a 30B agentic model, under an Apache 2.0 license on August 10, 2026. It runs on a single 24 GB consumer GPU, and Meta says open weights for its flagship Muse Spark 1.2 follow in the coming weeks.
Read the full story: Meta returns to open weights with Muse Glimmer 30B →
Transcript
Meta just opened the weights for Muse Glimmer, a thirty billion parameter agentic model, under a plain Apache two point zero license. It runs on a single twenty four gigabyte consumer GPU. At full precision it would need fifty five gigs, but Meta squeezed the weights under twenty, leaving room for the cache, the vision encoder and the drafter. Here's the part most coverage is missing. Meta published no SWE-bench score for it at all. And when Meta opens Muse Spark one point two in a few weeks, at eighty six point six percent, it lands third among downloadable models. Kimi K3 is at ninety three point four. DeepSeek V4 Flash, eighty eight point eight. Meta isn't leading open weights. It's rejoining a race that moved on.