| Benchmark | Gemini 3 Flash Preview | Opus 5 |
|---|---|---|
| AA Agentic Index | 49.7 | 59.2 |
| AA Intelligence | 38.7 | 63.1 |
| AA-LCR | 73 | 78.7 |
| AA-Omniscience | 10.1 | 37.1 |
| arc_agi_1 | 84.7 | 97.5 |
| ARC-AGI-2 | 33.6 | 90.4 |
| arena_elo | 1473 | 1487 |
| arena_text_factuality | 1452 | 1481 |
| Artificial Analysis Coding Index | 42.6 | 78 |
| coding_arena_elo | 1387 | 1662 |
| critpt | 8.6 | 29.1 |
| Finance Agent v2 | 42.6 | 58.6 |
| gdpval | 35.2 | 67.2 |
| GDPval-AA (Elo) | 1204 | 1827 |
| GPQA Diamond | 90.4 | 93.7 |
| HLE | 43.5 | 64.7 |
| Legal Agent Benchmark | 0 | 23 |
| MCP Atlas | 62 | 85.8 |
| MMMU-Pro | 81.2 | 84.7 |
| OmniScience Accuracy | 53.4 | 60.9 |
| OmniScience Non-Hallucination | 7.6 | 40.5 |
| OSWorld-Verified | 65.1 | 4 |
| scicode | 50.6 | 55.7 |
| simplebench | 61.1 | 80.6 |
| simpleqa_verified | 67.4 | 56.7 |
| SWE-bench Multilingual | 72.7 | 89.5 |
| SWE-bench Pro | 49.6 | 79.2 |
| SWE-bench Verified | 78 | 96 |
| TauBench V3 - Banking | 20.8 | 44.7 |
| Terminal-Bench 2.1 | 58 | 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.