Across 15 shared benchmarks, muse-glimmer-30b scores higher on 7 and Qwen3.6 27B on 8. The widest gap is baby_vision, where muse-glimmer-30b scores 70.4 against 28.9. muse-glimmer-30b is the cheaper of the two on tracked API pricing ($0.30 against $0.60 per million input tokens).
| Benchmark | muse-glimmer-30b | Qwen3.6 27B |
|---|---|---|
| AA-LCR | 80 | 73.3 |
| aime_2026 | 94.7 | 94.1 |
| baby_vision | 70.4 | 28.9 |
| charxiv_rq | 85.9 | 78.4 |
| GPQA Diamond | 83.5 | 87.8 |
| HLE | 22 | 24 |
| IFBench | 77 | 69.1 |
| MMMU-Pro | 74 | 75.8 |
| OmniDocBench 1.5 | 75.8 | 89.4 |
| OSWorld-Verified | 65.9 | 63.9 |
| scicode | 43.6 | 39.8 |
| SWE-bench Pro | 51.2 | 53.5 |
| SWE-bench Verified | 76 | 77.2 |
| Terminal Bench 2.1 (Terminus) | 51.7 | 63.4 |
| Terminal-Bench 2.1 | 51.7 | 60.7 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (deepinfra, direct), otherwise the lowest tracked offer.