| Benchmark | GPT-5 | o4-mini |
|---|---|---|
| AA Agentic Index | 49.7 | 36.1 |
| AA Intelligence | 35.3 | 26.1 |
| AA-LCR | 76.3 | 60 |
| AA-Omniscience | -8.7 | -35.7 |
| aider_polyglot | 88 | 72 |
| AIME 2025 | 95 | 92.7 |
| AIR-Bench 2024 | 87.7 | 78.5 |
| arc_agi_1 | 56.2 | 58.7 |
| ARC-AGI-2 | 7.5 | 6.1 |
| arena_vision | 1212 | 1201 |
| Artificial Analysis Coding Index | 38.9 | 25.6 |
| browsecomp | 54.9 | 51.5 |
| critpt | 5.7 | 0.6 |
| Fortress | 17 | 21.5 |
| frontiermath_tier_4 | 12.5 | 6.3 |
| gdpval | 32.5 | 25.4 |
| GPQA Diamond | 87.3 | 81.4 |
| GSM8K | -1.8 | 0 |
| harmbench | 70 | 97 |
| HLE | 28.5 | 16.5 |
| IFBench | 73.1 | 68.7 |
| imo_2025 | 38.1 | 14.3 |
| LiveCodeBench | 85 | 84.5 |
| MMMU | 84.2 | 81.6 |
| MMMU-Pro | 78.4 | 69.2 |
| multichallenge | 71.1 | 44.9 |
| MultiNRC | 52.1 | 22.2 |
| OmniScience Accuracy | 40.3 | 24.8 |
| OmniScience Non-Hallucination | 21 | 19.5 |
| scicode | 43 | 46.5 |
| SEAL VISTA | 49.7 | 51.8 |
| simplebench | 56.7 | 38.7 |
| simpleqa_verified | 50.6 | 23.9 |
| swe_bench_bash | 65 | 45 |
| SWE-bench Verified | 74.9 | 68.1 |
| Terminal-Bench Hard | 37.9 | 15.2 |
| vectara_answer_rate | 99.9 | 99.2 |
| vectara_avg_summary_length | 162.7 | 130.9 |
| vectara_factual_consistency | 85.3 | 81.4 |
| vectara_hallucination_rate ↓ | 14.7 | 18.6 |
| VisualToolBench | 18.7 | 11.1 |
| τ²-Bench Telecom (AA run) | 86.5 | 55.6 |
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.