Across 17 shared benchmarks, Claude Opus 4.6 scores higher on 12 and muse-glimmer-30b on 5. The widest gap is baby_vision, where muse-glimmer-30b scores 70.4 against 14.8. muse-glimmer-30b is the cheaper of the two on tracked API pricing ($0.30 against $5.00 per million input tokens).
| Benchmark | Claude Opus 4.6 | muse-glimmer-30b |
|---|---|---|
| AA-LCR | 74.3 | 80 |
| aime_2026 | 96.7 | 94.7 |
| baby_vision | 14.8 | 70.4 |
| charxiv_rq | 69.1 | 85.9 |
| deepsearchqa_f1 | 91.3 | 74.6 |
| GPQA Diamond | 91.3 | 83.5 |
| HLE | 53.1 | 22 |
| IFBench | 62.5 | 77 |
| MCP Atlas | 76.8 | 75.5 |
| MMMU-Pro | 77.3 | 74 |
| OmniDocBench 1.5 | 86.6 | 75.8 |
| OSWorld-Verified | 72.7 | 65.9 |
| scicode | 52 | 43.6 |
| screenspot_pro_no_tools | 57.7 | 75.4 |
| SWE-bench Pro | 57.3 | 51.2 |
| SWE-bench Verified | 80.8 | 76 |
| Terminal Bench 2.1 (Terminus) | 78.2 | 51.7 |
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 (anthropic-official, deepinfra), otherwise the lowest tracked offer.