| Benchmark | Muse Spark | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 28.7 | 14.4 |
| AA Intelligence | 44.3 | 7.7 |
| AA-LCR | 77 | 23 |
| AA-Omniscience | 7.2 | -68.4 |
| Artificial Analysis Coding Index | 58.6 | 6.7 |
| critpt | 11.3 | 0 |
| ERQA | 64.7 | 47.3 |
| gdpval | 32.3 | 13.8 |
| GPQA Diamond | 89.5 | 64.1 |
| HLE | 58.4 | 4.6 |
| IFBench | 75.9 | 36.6 |
| MMMU-Pro | 80.5 | 57 |
| OmniScience Accuracy | 49.6 | 12.2 |
| OmniScience Non-Hallucination | 26.8 | 8.2 |
| scicode | 51.5 | 17.1 |
| screenspot_pro_no_tools | 84.1 | 49.2 |
| Terminal-Bench Hard | 45.5 | 1.5 |
| τ²-Bench Telecom (AA run) | 91.5 | 15.5 |
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.