| Benchmark | Claude Sonnet 5 | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 49.7 | 70.4 |
| AA Intelligence | 55.3 | 52 |
| AA-LCR | 77 | 81 |
| AA-Omniscience | 16.4 | 21.2 |
| arena_elo | 1463 | 1477 |
| arena_text_factuality | 1453 | 1466 |
| Artificial Analysis Coding Index | 71.5 | 70.1 |
| charxiv_reasoning_with_tools | 88.3 | 84.9 |
| charxiv_rq | 77 | 84.2 |
| coding_arena_elo | 1540 | 1490 |
| critpt | 16.9 | 13.1 |
| DeepSWE 1.1 | 54 | 37 |
| gdpval | 54.8 | 57.8 |
| GDPval-AA v2 | 1618 | 1357 |
| GDPVal-AA v2 (Elo) | 1598 | 1349 |
| GPQA Diamond | 91.1 | 92.2 |
| HLE | 57.4 | 42.7 |
| Legal Agent Benchmark | 5.8 | 0.8 |
| livebench_agentic_coding | 59.4 | 49 |
| livebench_coding | 80.7 | 78.2 |
| livebench_data_analysis | 71.7 | 64.9 |
| livebench_instruction_following | 63.9 | 75.6 |
| livebench_language | 75 | 84.6 |
| livebench_math | 92.9 | 88.2 |
| livebench_reasoning | 88.7 | 82 |
| MLE-Bench | 66.9 | 49.7 |
| MMMU-Pro | 77.3 | 84.3 |
| mrcr_v2_8needle_128k_average | 81.5 | 77.3 |
| OmniScience Accuracy | 40 | 51.4 |
| OmniScience Non-Hallucination | 60.6 | 38.2 |
| OSWorld-Verified | 81.2 | 78.4 |
| scicode | 53.6 | 53.1 |
| simplebench | 60.6 | 76.7 |
| simpleqa_verified | 25 | 68.4 |
| SWE-bench Pro | 63.2 | 55.1 |
| TauBench V3 - Banking | 37.3 | 32.2 |
| Terminal-Bench 2.0 | 80.4 | 76.2 |
| Terminal-Bench 2.1 | 80.5 | 78.7 |
| toolathlon | 54.3 | 56.5 |
| τ³-Bench | 28.2 | 25.4 |
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.