| Benchmark | Gemma 4 26B A4B | Muse Spark |
|---|---|---|
| AA Agentic Index | 32.1 | 28.7 |
| AA Intelligence | 26.1 | 44.3 |
| AA-LCR | 61.7 | 77 |
| AA-Omniscience | -50.8 | 7.2 |
| arena_elo | 1438 | 1488 |
| arena_text_factuality | 1445 | 1467 |
| arena_vision | 1238 | 1295 |
| Artificial Analysis Coding Index | 39.3 | 58.6 |
| coding_arena_elo | 1361 | 1540 |
| critpt | 0 | 11.3 |
| gdpval | 25.7 | 32.3 |
| GPQA Diamond | 82.3 | 89.5 |
| HLE | 19.3 | 58.4 |
| IFBench | 72.4 | 75.9 |
| itbenchSre | 23.6 | 16.2 |
| MCP Atlas | 50 | 82.2 |
| MMMU-Pro | 73.8 | 80.5 |
| OmniScience Accuracy | 19.1 | 49.6 |
| OmniScience Non-Hallucination | 13.6 | 26.8 |
| scicode | 40.3 | 51.5 |
| SimpleVQA | 52.2 | 0.7 |
| SWE-bench Pro | 13.8 | 55 |
| SWE-bench Verified | 57.4 | 77.4 |
| TauBench V3 - Banking | 12 | 19.6 |
| Terminal-Bench 2.0 | 34.2 | 59 |
| Terminal-Bench 2.1 | 39 | 62.2 |
| Terminal-Bench Hard | 25 | 45.5 |
| τ²-Bench Telecom (AA run) | 43.6 | 91.5 |
| τ³-Bench | 59 | 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.