| Benchmark | Claude Sonnet 4.5 | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 50.6 | 18.8 |
| AA Intelligence | 37.4 | 17.7 |
| AA-LCR | 68.3 | 65.7 |
| AA-Omniscience | -0.1 | -29.8 |
| AIME 2025 | 88 | 72 |
| AIR-Bench 2024 | 89.8 | 68.7 |
| arc_agi_1 | 63.7 | 33.3 |
| ARC-AGI-2 | 13.6 | 2.5 |
| arena_text_factuality | 1448 | 1437 |
| Artificial Analysis Coding Index | 52.1 | 22.2 |
| critpt | 1.1 | 1.4 |
| frontiermath_tier_4 | 4.2 | 4.2 |
| gdpval | 40.4 | 11.9 |
| GPQA Diamond | 83.4 | 82.8 |
| HLE | 19.8 | 12.1 |
| hmmt_feb_2025 | 79.2 | 64.2 |
| IFBench | 57.3 | 50.3 |
| LiveCodeBench | 71 | 76.2 |
| MMMU | 79.6 | 79.7 |
| MMMU-Pro | 68.7 | 69.1 |
| mrcr | 55.4 | 32 |
| OmniScience Accuracy | 32.9 | 26.1 |
| OmniScience Non-Hallucination | 50.9 | 24.7 |
| scicode | 45 | 35.9 |
| SEAL VISTA | 48.8 | 49.1 |
| simplebench | 54.3 | 41.2 |
| swe_bench_bash | 71.4 | 28.7 |
| SWE-bench Verified | 82 | 60.4 |
| Terminal-Bench Hard | 35.6 | 13.6 |
| vectara_answer_rate | 95.6 | 99 |
| vectara_avg_summary_length | 127.8 | 101.5 |
| vectara_factual_consistency | 88 | 92.2 |
| vectara_hallucination_rate ↓ | 12 | 7.8 |
| VisualToolBench | 6.2 | 4.7 |
| τ²-Bench Telecom (AA run) | 78.1 | 31.6 |
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.