| Benchmark | Gemini 3 Pro | Opus 5 |
|---|---|---|
| AA Agentic Index | 52 | 59.2 |
| AA Intelligence | 40.6 | 63.1 |
| AA-LCR | 73 | 78.7 |
| AA-Omniscience | 15.3 | 37.1 |
| arc_agi_1 | 75 | 97.5 |
| ARC-AGI-2 | 54 | 90.4 |
| arena_elo | 1485 | 1487 |
| arena_text_factuality | 1481 | 1481 |
| Artificial Analysis Coding Index | 46.5 | 78 |
| browsecomp | 59.2 | 90.8 |
| coding_arena_elo | 1438 | 1662 |
| critpt | 9.1 | 29.1 |
| gdpval | 34.2 | 67.2 |
| GDPval-AA (Elo) | 1195 | 1827 |
| GPQA Diamond | 91.9 | 93.7 |
| HLE | 45.8 | 64.7 |
| HLE (with tools) | 45.8 | 64.7 |
| MCP Atlas | 70.3 | 85.8 |
| MMMU-Pro | 81 | 84.7 |
| OmniScience Accuracy | 55.8 | 60.9 |
| OmniScience Non-Hallucination | 10 | 40.5 |
| scicode | 56.1 | 55.7 |
| simplebench | 76.4 | 80.6 |
| simpleqa_verified | 72.9 | 56.7 |
| SWE-bench Multilingual | 68.7 | 89.5 |
| SWE-bench Pro | 43.3 | 79.2 |
| SWE-bench Verified | 78 | 96 |
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.