| Benchmark | Claude 4 Opus | GPT-5 |
|---|---|---|
| AA Intelligence | 31.7 | 35.3 |
| AA-LCR | 40 | 76.3 |
| aider_polyglot | 70.7 | 88 |
| AIME | 33.9 | 7.6 |
| AIME 2025 | 75.5 | 95 |
| AIR-Bench 2024 | 85.7 | 87.7 |
| arc_agi_1 | 27 | 56.2 |
| ARC-AGI-2 | 8.6 | 7.5 |
| arena_vision | 1207 | 1212 |
| Artificial Analysis Coding Index | 34 | 38.9 |
| Fortress | 27.6 | 17 |
| frontiermath_tier_4 | 4.2 | 12.5 |
| GPQA Diamond | 79.6 | 87.3 |
| harmbench | 91.7 | 70 |
| HLE | 12.3 | 28.5 |
| HMMT 2025 | 15.9 | 93.3 |
| hmmt_feb_2025 | 60 | 88.3 |
| IFBench | 53.7 | 73.1 |
| LiveCodeBench | 70.4 | 85 |
| LiveCodeBench v6 | 47.4 | 87 |
| metr_hcast | 61.7 | 68.4 |
| metr_re_bench | 20 | 40 |
| metr_swaa | 99.8 | 100 |
| mmlu_redux | 94.2 | 95.3 |
| MMLU-Pro | 86.6 | 87.5 |
| MMMU | 73.7 | 84.2 |
| multichallenge | 58.6 | 71.1 |
| MultiNRC | 33.9 | 52.1 |
| scicode | 40.9 | 43 |
| SEAL VISTA | 47 | 49.7 |
| simplebench | 58.8 | 56.7 |
| swe_bench_bash | 67.6 | 65 |
| SWE-bench Verified | 79.4 | 74.9 |
| Tau2 airline | 60 | 62.6 |
| Tau2 retail | 81.8 | 81.1 |
| Terminal-Bench | 43.2 | 43.8 |
| Terminal-Bench Hard | 31.1 | 37.9 |
| vectara_answer_rate | 91 | 99.9 |
| vectara_avg_summary_length | 123.2 | 162.7 |
| vectara_factual_consistency | 88 | 85.3 |
| vectara_hallucination_rate ↓ | 12 | 14.7 |
| τ²-Bench | 57 | 85 |
| τ²-Bench Telecom (AA run) | 73.4 | 86.5 |
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.