| Benchmark | Gemini 2.5 Pro | Opus 5 |
|---|---|---|
| AA Agentic Index | 7.2 | 59.2 |
| AA Intelligence | 27 | 63.1 |
| AA-LCR | 66 | 78.7 |
| AA-Omniscience | -14.3 | 37.1 |
| arc_agi_1 | 41 | 97.5 |
| ARC-AGI-2 | 4.9 | 90.4 |
| arena_elo | 1446 | 1487 |
| arena_text_factuality | 1462 | 1481 |
| Artificial Analysis Coding Index | 46.7 | 78 |
| browsecomp | 9.9 | 90.8 |
| coding_arena_elo | 1224 | 1662 |
| critpt | 2.6 | 29.1 |
| gdpval | 8.5 | 67.2 |
| GPQA Diamond | 84.4 | 93.7 |
| HLE | 22.5 | 64.7 |
| HLE (with tools) | 28.4 | 64.7 |
| MMMU-Pro | 74.9 | 84.7 |
| OmniScience Accuracy | 39 | 60.9 |
| OmniScience Non-Hallucination | 12.6 | 40.5 |
| scicode | 43 | 55.7 |
| simplebench | 62.4 | 80.6 |
| simpleqa_verified | 56 | 56.7 |
| SWE-bench Verified | 67.2 | 96 |
| TauBench V3 - Banking | 9.7 | 44.7 |
| Terminal-Bench 2.1 | 28.5 | 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.