| Benchmark | Claude Sonnet 5 | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 49.7 | 32.1 |
| AA Intelligence | 55.3 | 26.1 |
| AA-LCR | 77 | 61.7 |
| AA-Omniscience | 16.4 | -50.8 |
| arena_elo | 1463 | 1438 |
| arena_text_factuality | 1453 | 1445 |
| Artificial Analysis Coding Index | 71.5 | 39.3 |
| browsecomp | 84.7 | 26.3 |
| charxiv_rq | 77 | 69 |
| coding_arena_elo | 1540 | 1361 |
| critpt | 16.9 | 0 |
| gdpval | 54.8 | 25.7 |
| GDPval-AA v2 | 1618 | 807 |
| GPQA Diamond | 91.1 | 82.3 |
| HLE | 57.4 | 19.3 |
| MMMU-Pro | 77.3 | 73.8 |
| mrcr_v2_8needle_128k_average | 81.5 | 44.1 |
| OmniScience Accuracy | 40 | 19.1 |
| OmniScience Non-Hallucination | 60.6 | 13.6 |
| scicode | 53.6 | 40.3 |
| SWE-bench Multilingual | 78.3 | 43.4 |
| SWE-bench Pro | 63.2 | 13.8 |
| SWE-bench Verified | 85.2 | 57.4 |
| TauBench V3 - Banking | 37.3 | 12 |
| Terminal-Bench 2.0 | 80.4 | 34.2 |
| Terminal-Bench 2.1 | 80.5 | 39 |
| τ³-Bench | 28.2 | 59 |
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.