| Benchmark | Gemma 4 31B | Opus 5 |
|---|---|---|
| AA Agentic Index | 14.4 | 59.2 |
| AA Intelligence | 29.7 | 63.1 |
| AA-LCR | 68.3 | 78.7 |
| AA-Omniscience | -47.9 | 37.1 |
| arena_elo | 1451 | 1487 |
| arena_text_factuality | 1449 | 1481 |
| Artificial Analysis Coding Index | 43.4 | 78 |
| coding_arena_elo | 1364 | 1662 |
| critpt | 1.4 | 29.1 |
| gdpval | 15.5 | 67.2 |
| GPQA Diamond | 85.7 | 93.7 |
| HLE | 26.5 | 64.7 |
| HLE (with tools) | 26.5 | 64.7 |
| MCP Atlas | 57.2 | 85.8 |
| MMMU-Pro | 76.9 | 84.7 |
| OmniScience Accuracy | 20 | 60.9 |
| OmniScience Non-Hallucination | 18.1 | 40.5 |
| scicode | 43.4 | 55.7 |
| simpleqa_verified | 9.6 | 56.7 |
| SWE-bench Multilingual | 51.7 | 89.5 |
| SWE-bench Pro | 35.7 | 79.2 |
| SWE-bench Verified | 52 | 96 |
| TauBench V3 - Banking | 14.8 | 44.7 |
| Terminal-Bench 2.1 | 43.4 | 89.1 |
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.