| Benchmark | Claude Sonnet 5 | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 49.7 | 40.5 |
| AA Intelligence | 55.3 | 52 |
| AA-LCR | 77 | 79 |
| AA-Omniscience | 16.4 | 22.1 |
| Agent's Last Exam (Pass rate) | 33.3 | 24.2 |
| arena_elo | 1463 | 1481 |
| arena_text_factuality | 1453 | 1470 |
| Artificial Analysis Coding Index | 71.5 | 69.2 |
| charxiv_reasoning_with_tools | 88.3 | 89.4 |
| charxiv_rq | 77 | 85.2 |
| Code Arena (Elo) | 1541 | 1538 |
| coding_arena_elo | 1540 | 1538 |
| critpt | 16.9 | 10.6 |
| DeepSWE 1.1 | 54 | 49 |
| GDP (Surge AI) | 81.6 | 22 |
| gdpval | 54.8 | 46.1 |
| GDPval-AA v2 | 1618 | 1421 |
| GDPVal-AA v2 (Elo) | 1598 | 1422 |
| GPQA Diamond | 91.1 | 92.8 |
| HLE | 57.4 | 40.8 |
| HLE-Verified | 31 | 51.2 |
| livebench_agentic_coding | 59.4 | 43.4 |
| livebench_coding | 80.7 | 77.9 |
| livebench_data_analysis | 71.7 | 63 |
| livebench_instruction_following | 63.9 | 75.4 |
| livebench_language | 75 | 83.9 |
| livebench_math | 92.9 | 86.4 |
| livebench_reasoning | 88.7 | 85.2 |
| LVBench | 68.5 | 84.2 |
| MLE-Bench | 66.9 | 63.9 |
| MMMU-Pro | 77.3 | 83.2 |
| mrcr_v2_8needle_128k_average | 81.5 | 91.8 |
| OmniScience Accuracy | 40 | 50 |
| OmniScience Non-Hallucination | 60.6 | 44.4 |
| OSWorld-Verified | 81.2 | 83 |
| scicode | 53.6 | 52.7 |
| simpleqa_verified | 25 | 68.7 |
| SWE-bench Pro | 63.2 | 58.7 |
| TauBench V3 - Banking | 37.3 | 29.9 |
| Terminal-Bench 2.1 | 80.5 | 78 |
| Terminal-bench 3.0 | 14.6 | 5.4 |
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.