| Benchmark | Muse Spark | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 28.7 | 54.6 |
| AA Intelligence | 44.3 | 34.6 |
| AA-LCR | 77 | 72.3 |
| AA-Omniscience | 7.2 | -44 |
| arena_vision | 1295 | 1221 |
| Artificial Analysis Coding Index | 58.6 | 34.9 |
| coding_arena_elo | 1540 | 1357 |
| critpt | 11.3 | 0.9 |
| ERQA | 64.7 | 60.5 |
| gdpval | 32.3 | 33 |
| GPQA Diamond | 89.5 | 85.8 |
| HLE | 58.4 | 48.5 |
| IFBench | 75.9 | 76.5 |
| itbenchSre | 16.2 | 35.5 |
| MCP Atlas | 82.2 | 68.4 |
| MMMU-Pro | 80.5 | 75 |
| multichallenge | 75.5 | 60.8 |
| OmniScience Accuracy | 49.6 | 20.7 |
| OmniScience Non-Hallucination | 26.8 | 24.9 |
| scicode | 51.5 | 39.5 |
| screenspot_pro_no_tools | 84.1 | 70.3 |
| SimpleVQA | 0.7 | 56 |
| SWE-bench Pro | 55 | 51.2 |
| SWE-bench Verified | 77.4 | 75 |
| Terminal-Bench 2.0 | 59 | 41.6 |
| Terminal-Bench Hard | 45.5 | 32.6 |
| ZeroBench | 0.3 | 0.1 |
| τ²-Bench Telecom (AA run) | 91.5 | 93.9 |
| τ³-Bench | 19.6 | 68.4 |
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.