Muse Glimmer 30B (DeepInfra)

Meta Superintelligence Labs released Muse Glimmer on August 10, 2026, as a dense open-weight model with approximately 30 billion parameters (29.6 billion by Gigazine's count), positioned by Meta as small enough to run on a single consumer GPU without cloud or metered API dependencies. The page describes Glimmer as a di The piece situates Muse Glimmer as Meta's first open-weight model drop since Llama 4, arriving at a time when open-weight releases from Qwen, DeepSeek, and Kimi have been setting the pace in raw parameter count. It frames the release as Meta's attempt to reclaim ground in the open-weight race, targeting developers who

Third-party benchmark telemetry from llm-benchmarks.com measures Muse Glimmer 30B on the DeepInfra lane under catalog id "metamodelsmuseglimmer30b," reporting an average throughput of 15.70 tokens per second and an average time to first token of 3600.00 ms across 3 recent runs. The page also records a decode throughput The benchmark page provides current performance data for the DeepInfra-hosted Muse Glimmer 30B variant referenced in the subject, with sample sizes and update cadence clearly disclosed. With only 3 runs analyzed, the figures carry statistical limitations, but they constitute the most concrete direct evidence in the sup