| Benchmark | Claude Opus 4.7 | GPT-4o |
|---|---|---|
| AA Agentic Index | 64.6 | 9.7 |
| AA Intelligence | 55 | 11.2 |
| AA-LCR | 75.3 | 55 |
| AA-Omniscience | 27.3 | -10.5 |
| arc_agi_1 | 92 | 4.5 |
| ARC-AGI-2 | 75.8 | 0 |
| arena_vision | 1304 | 1162 |
| Artificial Analysis Coding Index | 73.6 | 24.2 |
| critpt | 12 | 0 |
| gaia_overall | 28 | 28.9 |
| gdpval | 58.6 | 0 |
| GPQA Diamond | 94.2 | 70.1 |
| harmbench | 17.5 | 82.9 |
| HLE | 54.7 | 5.3 |
| IFBench | 58.6 | 36 |
| mmmlu | 91.5 | 81.4 |
| MMMU-Pro | 78.8 | 59.9 |
| OmniScience Accuracy | 48.9 | 23.7 |
| OmniScience Non-Hallucination | 57.7 | 62.1 |
| OSWorld-Verified | 82.8 | 5 |
| scicode | 54.5 | 33.4 |
| simplebench | 61.7 | 17.8 |
| SWE-bench Verified | 87.6 | 38.8 |
| Terminal-Bench Hard | 54.5 | 8.3 |
| vectara_answer_rate | 98 | 93.8 |
| vectara_avg_summary_length | 149.1 | 86.6 |
| vectara_factual_consistency | 88 | 90.4 |
| vectara_hallucination_rate ↓ | 12 | 9.6 |
| τ²-Bench Telecom (AA run) | 88.6 | 28.9 |
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.