| Benchmark | Claude Opus 4.5 | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 59.6 | 18.8 |
| AA Intelligence | 41.9 | 17.7 |
| AA-LCR | 76 | 65.7 |
| AA-Omniscience | 14 | -29.8 |
| AIME 2025 | 92.8 | 72 |
| arc_agi_1 | 82 | 33.3 |
| ARC-AGI-2 | 37.6 | 2.5 |
| arena_text_factuality | 1458 | 1437 |
| Artificial Analysis Coding Index | 47.8 | 22.2 |
| critpt | 4.6 | 1.4 |
| frontiermath_tier_4 | 4.2 | 4.2 |
| gdpval | 47.3 | 11.9 |
| GPQA Diamond | 87 | 82.8 |
| HLE | 30.8 | 12.1 |
| hmmt_feb_2025 | 92.9 | 64.2 |
| IFBench | 58 | 50.3 |
| LiveCodeBench | 87 | 76.2 |
| MMMU | 80.7 | 79.7 |
| MMMU-Pro | 74 | 69.1 |
| OmniScience Accuracy | 46.6 | 26.1 |
| OmniScience Non-Hallucination | 39 | 24.7 |
| scicode | 50 | 35.9 |
| SEAL VISTA | 46.4 | 49.1 |
| simplebench | 62 | 41.2 |
| swe_bench_bash | 76.8 | 28.7 |
| SWE-bench Verified | 81.5 | 60.4 |
| Terminal-Bench Hard | 47 | 13.6 |
| vectara_answer_rate | 98.7 | 99 |
| vectara_avg_summary_length | 114.5 | 101.5 |
| vectara_factual_consistency | 89.1 | 92.2 |
| vectara_hallucination_rate ↓ | 10.9 | 7.8 |
| τ²-Bench Telecom (AA run) | 89.5 | 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.