Muse Glimmer 30B: An On-Device Agentic Model

Benchmark Model Rank Results
agents-on-mcpatlas-public-500-tasksMuse Glimmer-30B (High Reasoning)#5Accuracy: 75.5
chart-question-answering-on-charxiv-rqMuse Glimmer-30B (High Reasoning)#5Accuracy: 78.8
instruction-following-on-ifbenchMuse Glimmer-30B (High Reasoning)#8Accuracy: 77.0
mathematical-reasoning-on-aime-2026Muse Glimmer-30B (High Reasoning)#9Accuracy: 94.7
code-generation-on-scicodeMuse Glimmer-30B (High)#11Accuracy: 43.6
multimodal-reasoning-on-mmmu-proMuse Glimmer-30B (High Reasoning)#14Accuracy: 74.0
code-generation-on-swe-bench-verifiedMuse Glimmer-30B (High)#22Accuracy: 76.0
question-answering-on-humanity-s-last-exam-hleMuse Glimmer-30B (High Reasoning, text-only)#25Accuracy: 22.0
code-generation-on-swe-bench-proMuse Glimmer-30B (High)#31Resolved: 51.2
code-generation-on-terminal-bench-2-1Muse Glimmer-30B (High) (Terminus 2)#39Accuracy: 51.7
multiple-choice-question-answering-mcqa-on-gpqa-diamondMuse Glimmer-30B (High Reasoning)#45Accuracy: 83.5