| Benchmark | Claude Opus 4.8 | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 49.4 | 18.8 |
| AA Intelligence | 57.3 | 17.7 |
| AA-LCR | 73 | 65.7 |
| AA-Omniscience | 28.8 | -29.8 |
| arc_agi_1 | 92.5 | 33.3 |
| ARC-AGI-2 | 72.1 | 2.5 |
| arena_text_factuality | 1464 | 1437 |
| arena_vision | 1280 | 1214 |
| Artificial Analysis Coding Index | 74.3 | 22.2 |
| critpt | 20.9 | 1.4 |
| frontiermath_tier_4 | 31.3 | 4.2 |
| gdpval | 54.2 | 11.9 |
| GPQA Diamond | 93.6 | 82.8 |
| HLE | 57.9 | 12.1 |
| IFBench | 62.2 | 50.3 |
| MMMU-Pro | 78.9 | 69.1 |
| OmniScience Accuracy | 48.8 | 26.1 |
| OmniScience Non-Hallucination | 60.7 | 24.7 |
| scicode | 53.5 | 35.9 |
| simplebench | 64.8 | 41.2 |
| SWE-bench Verified | 88.6 | 60.4 |
| Terminal-Bench Hard | 58.3 | 13.6 |
| τ²-Bench Telecom (AA run) | 94.4 | 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.