| Benchmark | Gemini 3.5 Flash | Opus 5 |
|---|---|---|
| AA Agentic Index | 70.4 | 59.2 |
| AA Intelligence | 52 | 63.1 |
| AA-LCR | 81 | 78.7 |
| AA-Omniscience | 21.2 | 37.1 |
| arc_agi_1 | 92.5 | 97.5 |
| ARC-AGI-2 | 72.1 | 90.4 |
| arena_elo | 1477 | 1487 |
| arena_text_factuality | 1466 | 1481 |
| Artificial Analysis Coding Index | 70.1 | 78 |
| coding_arena_elo | 1490 | 1662 |
| critpt | 13.1 | 29.1 |
| DeepSWE 1.1 | 37 | 68.8 |
| Finance Agent v2 | 57.9 | 58.6 |
| gdpval | 57.8 | 67.2 |
| GDPval-AA (Elo) | 1656 | 1827 |
| GDPval-AA v2 | 1357 | 1861 |
| GPQA Diamond | 92.2 | 93.7 |
| HiL-Bench | 27.7 | 57 |
| HLE | 42.7 | 64.7 |
| Legal Agent Benchmark | 0.8 | 23 |
| livebench_agentic_coding | 49 | 65.2 |
| livebench_coding | 78.2 | 81.5 |
| livebench_data_analysis | 64.9 | 74.5 |
| livebench_instruction_following | 75.6 | 63.8 |
| livebench_language | 84.6 | 88.7 |
| livebench_math | 88.2 | 95.7 |
| livebench_reasoning | 82 | 91.2 |
| MCP Atlas | 83.6 | 85.8 |
| MMMU-Pro | 84.3 | 84.7 |
| OmniScience Accuracy | 51.4 | 60.9 |
| OmniScience Non-Hallucination | 38.2 | 40.5 |
| OSWorld-Verified | 78.4 | 4 |
| scicode | 53.1 | 55.7 |
| simplebench | 76.7 | 80.6 |
| simpleqa_verified | 68.4 | 56.7 |
| SWE-bench Pro | 55.1 | 79.2 |
| TauBench V3 - Banking | 32.2 | 44.7 |
| Terminal-Bench 2.1 | 78.7 | 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.