Across 16 shared benchmarks, muse-glimmer-30b scores higher on 12 and Qwen3.5 35B A3B on 4. The widest gap is HLE, where Qwen3.5 35B A3B scores 47.4 against 22. Tracked API pricing per million tokens: muse-glimmer-30b $0.30 in / $1.20 out, Qwen3.5 35B A3B $0.25 in / $2.00 out.
| Benchmark | muse-glimmer-30b | Qwen3.5 35B A3B |
|---|---|---|
| AA-LCR | 80 | 68.3 |
| aime_2026 | 94.7 | 93.3 |
| baby_vision | 70.4 | 38.4 |
| charxiv_rq | 85.9 | 77.5 |
| GPQA Diamond | 83.5 | 84.5 |
| HLE | 22 | 47.4 |
| IFBench | 77 | 72.5 |
| MCP Atlas | 75.5 | 62.4 |
| MMMU-Pro | 74 | 75.1 |
| OmniDocBench 1.5 | 75.8 | 89.3 |
| OSWorld-Verified | 65.9 | 54.5 |
| scicode | 43.6 | 37.7 |
| screenspot_pro_no_tools | 75.4 | 68.6 |
| SWE-bench Pro | 51.2 | 44.6 |
| SWE-bench Verified | 76 | 70 |
| Terminal-Bench 2.1 | 51.7 | 40.8 |
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.