Meta's Superintelligence Labs has released Muse Glimmer, a 30B-parameter model designed to run as an always-on local agent on a single consumer GPU, with weights available on Hugging Face under an Apache 2.0 license for commercial use. In benchmarks Meta ran, Muse Glimmer outperformed both Gemma4-31B and Qwen3.6-27B on most tests. On MCP Atlas, Muse Glimmer scored 75.5 against Gemma4-31B’s 54.2 and Qwen3.6-27B’s 62.5, with similar leads on DeepSearch QA (74.6 vs 61.7 and 71.1), WildClawBench (47.6 vs 37.6 and 43.2), and GAIA2 (43.3 vs 36.4 and 40.0). In coding, Muse Glimmer took SWE-Bench Pro at 51.2 against Gemma4-31B’s 36.9 and edged both rivals on SciCode with 43.6, but Qwen3.6-27B led on TerminalBench 2.1 (60.7 to 51.7), SWE-Bench Verified (77.2 to 76.0), and OSWorld-Verified (75.6 to 65.9). Qwen also held a slight edge on multimodal rows including ScreenSpot Pro,