| Benchmark | Claude Sonnet 5 | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 49.7 | 14.4 |
| AA Intelligence | 55.3 | 29.7 |
| AA-LCR | 77 | 68.3 |
| AA-Omniscience | 16.4 | -47.9 |
| arena_elo | 1463 | 1451 |
| arena_text_factuality | 1453 | 1449 |
| Artificial Analysis Coding Index | 71.5 | 43.4 |
| charxiv_rq | 77 | 67.9 |
| coding_arena_elo | 1540 | 1364 |
| critpt | 16.9 | 1.4 |
| gdpval | 54.8 | 15.5 |
| GPQA Diamond | 91.1 | 85.7 |
| HLE | 57.4 | 26.5 |
| MMMU-Pro | 77.3 | 76.9 |
| mrcr_v2_8needle_128k_average | 81.5 | 66.4 |
| OmniScience Accuracy | 40 | 20 |
| OmniScience Non-Hallucination | 60.6 | 18.1 |
| scicode | 53.6 | 43.4 |
| simpleqa_verified | 25 | 9.6 |
| SWE-bench Multilingual | 78.3 | 51.7 |
| SWE-bench Pro | 63.2 | 35.7 |
| SWE-bench Verified | 85.2 | 52 |
| TauBench V3 - Banking | 37.3 | 14.8 |
| Terminal-Bench 2.0 | 80.4 | 42.9 |
| Terminal-Bench 2.1 | 80.5 | 43.4 |
| τ³-Bench | 28.2 | 67.5 |
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.