| Benchmark | Gemma 4 31B | Muse Spark |
|---|---|---|
| AA Agentic Index | 14.4 | 28.7 |
| AA Intelligence | 29.7 | 44.3 |
| AA-LCR | 68.3 | 77 |
| AA-Omniscience | -47.9 | 7.2 |
| arena_elo | 1451 | 1488 |
| arena_text_factuality | 1449 | 1467 |
| arena_vision | 1252 | 1295 |
| Artificial Analysis Coding Index | 43.4 | 58.6 |
| coding_arena_elo | 1364 | 1540 |
| critpt | 1.4 | 11.3 |
| ERQA | 57.5 | 64.7 |
| gdpval | 15.5 | 32.3 |
| GPQA Diamond | 85.7 | 89.5 |
| HLE | 26.5 | 58.4 |
| IFBench | 75.6 | 75.9 |
| itbenchSre | 37.3 | 16.2 |
| MCP Atlas | 57.2 | 82.2 |
| MMMU-Pro | 76.9 | 80.5 |
| OmniScience Accuracy | 20 | 49.6 |
| OmniScience Non-Hallucination | 18.1 | 26.8 |
| scicode | 43.4 | 51.5 |
| simpleqa_verified | 9.6 | 66.3 |
| SimpleVQA | 52.9 | 0.7 |
| SWE-bench Pro | 35.7 | 55 |
| SWE-bench Verified | 52 | 77.4 |
| TauBench V3 - Banking | 14.8 | 19.6 |
| Terminal-Bench 2.0 | 42.9 | 59 |
| Terminal-Bench 2.1 | 43.4 | 62.2 |
| Terminal-Bench Hard | 36.4 | 45.5 |
| τ²-Bench Telecom (AA run) | 65.5 | 91.5 |
| τ³-Bench | 67.5 | 19.6 |
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.