| Benchmark | Claude Opus 4.6 | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 67.6 | 18.8 |
| AA Intelligence | 44.9 | 17.7 |
| AA-LCR | 74.3 | 65.7 |
| AA-Omniscience | 13.7 | -29.8 |
| AIME 2025 | 99.8 | 72 |
| arc_agi_1 | 93 | 33.3 |
| ARC-AGI-2 | 68.8 | 2.5 |
| arena_text_factuality | 1487 | 1437 |
| arena_vision | 1300 | 1214 |
| Artificial Analysis Coding Index | 48.1 | 22.2 |
| critpt | 12.6 | 1.4 |
| frontiermath_tier_4 | 22.9 | 4.2 |
| gdpval | 55.9 | 11.9 |
| GPQA Diamond | 91.3 | 82.8 |
| HLE | 53.1 | 12.1 |
| IFBench | 62.5 | 50.3 |
| LiveCodeBench | 88.8 | 76.2 |
| MMMU-Pro | 77.3 | 69.1 |
| OmniScience Accuracy | 47 | 26.1 |
| OmniScience Non-Hallucination | 37.2 | 24.7 |
| scicode | 52 | 35.9 |
| SEAL VISTA | 46.1 | 49.1 |
| simplebench | 67.6 | 41.2 |
| swe_bench_bash | 75.6 | 28.7 |
| SWE-bench Verified | 80.8 | 60.4 |
| Terminal-Bench Hard | 65.4 | 13.6 |
| vectara_answer_rate | 99.8 | 99 |
| vectara_avg_summary_length | 137.6 | 101.5 |
| vectara_factual_consistency | 87.8 | 92.2 |
| vectara_hallucination_rate ↓ | 12.2 | 7.8 |
| VisualToolBench | 27.5 | 4.7 |
| τ²-Bench Telecom (AA run) | 99.3 | 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.