| Benchmark | Gemini 3.6 Flash | Opus 5 |
|---|---|---|
| AA Agentic Index | 40.5 | 59.2 |
| AA Intelligence | 52 | 63.1 |
| AA-LCR | 79 | 78.7 |
| AA-Omniscience | 22.1 | 37.1 |
| arc_agi_1 | 34.5 | 97.5 |
| ARC-AGI-2 | 2.6 | 90.4 |
| arena_elo | 1481 | 1487 |
| arena_text_factuality | 1470 | 1481 |
| Artificial Analysis Coding Index | 69.2 | 78 |
| coding_arena_elo | 1538 | 1662 |
| critpt | 10.6 | 29.1 |
| DeepSWE 1.1 | 49 | 68.8 |
| GDP (Surge AI) | 22 | 85.5 |
| gdpval | 46.1 | 67.2 |
| GDPval-AA v2 | 1421 | 1861 |
| GPQA Diamond | 92.8 | 93.7 |
| HLE | 40.8 | 64.7 |
| livebench_agentic_coding | 43.4 | 65.2 |
| livebench_coding | 77.9 | 81.5 |
| livebench_data_analysis | 63 | 74.5 |
| livebench_instruction_following | 75.4 | 63.8 |
| livebench_language | 83.9 | 88.7 |
| livebench_math | 86.4 | 95.7 |
| livebench_reasoning | 85.2 | 91.2 |
| MMMU-Pro | 83.2 | 84.7 |
| OmniScience Accuracy | 50 | 60.9 |
| OmniScience Non-Hallucination | 44.4 | 40.5 |
| OSWorld 2.0 | 33.8 | 70.6 |
| OSWorld-Verified | 83 | 4 |
| scicode | 52.7 | 55.7 |
| simpleqa_verified | 68.7 | 56.7 |
| SWE-bench Pro | 58.7 | 79.2 |
| TauBench V3 - Banking | 29.9 | 44.7 |
| Terminal-Bench 2.1 | 78 | 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.