| Benchmark | Claude 4 Opus | GPT-5.5 |
|---|---|---|
| AA Intelligence | 31.7 | 56.3 |
| AA-LCR | 40 | 79 |
| arc_agi_1 | 27 | 95 |
| ARC-AGI-2 | 8.6 | 85 |
| arena_vision | 1207 | 1283 |
| Artificial Analysis Coding Index | 34 | 74.9 |
| Fortress | 27.6 | 16.3 |
| frontiermath_tier_4 | 4.2 | 35.4 |
| GPQA Diamond | 79.6 | 93.6 |
| HLE | 12.3 | 52.2 |
| IFBench | 53.7 | 75.9 |
| livebench | 74.6 | 80.7 |
| scicode | 40.9 | 56.1 |
| simplebench | 58.8 | 69 |
| SWE-bench Verified | 79.4 | 80.6 |
| Terminal-Bench Hard | 31.1 | 60.6 |
| vectara_answer_rate | 91 | 100 |
| vectara_avg_summary_length | 123.2 | 129.6 |
| vectara_factual_consistency | 88 | 90.7 |
| vectara_hallucination_rate ↓ | 12 | 9.3 |
| τ²-Bench Telecom (AA run) | 73.4 | 93.9 |
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.