| Benchmark | Claude 4 Opus | GPT-5.2 |
|---|---|---|
| AA Intelligence | 31.7 | 43.3 |
| AA-LCR | 40 | 79.3 |
| AIME 2025 | 75.5 | 100 |
| arc_agi_1 | 27 | 86.2 |
| ARC-AGI-2 | 8.6 | 52.9 |
| arena_vision | 1207 | 1245 |
| Artificial Analysis Coding Index | 34 | 48.7 |
| Fortress | 27.6 | 17.5 |
| frontiermath_tier_4 | 4.2 | 18.8 |
| GPQA Diamond | 79.6 | 92.4 |
| HLE | 12.3 | 45.5 |
| HMMT 2025 | 15.9 | 99.4 |
| hmmt_feb_2025 | 60 | 100 |
| IFBench | 53.7 | 75.4 |
| livebench | 74.6 | 74.8 |
| LiveCodeBench | 70.4 | 89 |
| longbench_v2 | 55.6 | 54.5 |
| metr_hcast | 61.7 | 77 |
| metr_re_bench | 20 | 35 |
| metr_swaa | 99.8 | 99.6 |
| MMLU-Pro | 86.6 | 87 |
| mmmlu | 88.8 | 89.6 |
| MultiNRC | 33.9 | 42.2 |
| scicode | 40.9 | 52.1 |
| SEAL VISTA | 47 | 46.6 |
| simplebench | 58.8 | 45.8 |
| swe_bench_bash | 67.6 | 72.8 |
| SWE-bench Verified | 79.4 | 80 |
| Tau2 retail | 81.8 | 82 |
| Terminal-Bench Hard | 31.1 | 62.2 |
| vectara_answer_rate | 91 | 100 |
| vectara_avg_summary_length | 123.2 | 186.3 |
| vectara_factual_consistency | 88 | 89.2 |
| vectara_hallucination_rate ↓ | 12 | 10.8 |
| τ²-Bench | 57 | 85.5 |
| τ²-Bench Telecom (AA run) | 73.4 | 98.7 |
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.