| Benchmark | Claude 4 Opus | GPT-5.4 |
|---|---|---|
| AA Intelligence | 31.7 | 53.1 |
| AA-LCR | 40 | 77.7 |
| arc_agi_1 | 27 | 93.7 |
| ARC-AGI-2 | 8.6 | 74 |
| arena_vision | 1207 | 1277 |
| Artificial Analysis Coding Index | 34 | 71.1 |
| frontiermath_tier_4 | 4.2 | 27.1 |
| GPQA Diamond | 79.6 | 93 |
| HLE | 12.3 | 43.7 |
| IFBench | 53.7 | 73.9 |
| livebench | 74.6 | 80.3 |
| MMLU-Pro | 86.6 | 87.5 |
| MultiNRC | 33.9 | 58.3 |
| scicode | 40.9 | 56.6 |
| SEAL VISTA | 47 | 50.9 |
| Terminal-Bench Hard | 31.1 | 57.6 |
| vectara_answer_rate | 91 | 99.9 |
| vectara_avg_summary_length | 123.2 | 81.7 |
| vectara_factual_consistency | 88 | 93 |
| vectara_hallucination_rate ↓ | 12 | 7 |
| τ²-Bench Telecom (AA run) | 73.4 | 87.1 |
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.