Muse Glimmer 30B: benchmark results
Meta's open-weight Muse Glimmer (August 2026), a 30-billion-parameter model distilled from Muse Spark for autonomous agentic tasks on consumer hardware, with a dedicated perception encoder for multimodal tool use. Provider: Meta. Released 2026-08-09. Access: Open.
Aggregate Rating 1708 ± 9, rank #178 of 1656 rated models, from 549 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AMIGO (Threshold 0.8) | 11.5 | Verified Correct accuracy in percent on AMIGO's dress galler | 100 |
| CPTU Bench | 4.42 | Average Score (1-5) | 100 |
| Jais 3 Model Cards - Arabic GSM8K (IFM-audited) | 95.2 | Accuracy (%) | 100 |
| Meta Muse Glimmer 30B Card - BEAM-128K | 65.1 | Mean Pass@1 (%) | 100 |
| Meta Muse Glimmer 30B Card - Siren AgentDojo - Utility | 94.2 | Utility Under Attack (%) | 100 |
| MetroLLM-Bench | 92.03 | Composite score (238 held-out cases, higher is better) | 100 |
| SEA-HELM (Burmese) - Belebele QA | 85.88 | Normalized Score | 100 |
| SEA-HELM (Burmese) - NLU | 67.54 | Normalized Score | 100 |
| SEA-HELM (English) - AA-LCR | 81.42 | Normalized Score | 100 |
| SEA-HELM (Indonesian) - Metaphor Understanding | 89.85 | Normalized Score | 100 |
| SEA-HELM (Indonesian) - NLU | 83.99 | Normalized Score | 100 |
| SEA-HELM (Indonesian) - SEA-Safeguard | 78.42 | Normalized Score | 100 |
Interactive version: theaggregate.ai/model?slug=muse-glimmer-30b · How It Works · Data refreshed daily, snapshot 2026-10-20.