GPT-5 (High): benchmark results
GPT-5 evaluated at the high reasoning-effort setting. Provider: OpenAI. Released 2025-08-07. Access: API.
Aggregate Rating 1743 ± 7, rank #254 of 2168 rated models, from 331 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA MATH-500 | 99.4 | Accuracy (%) | 100 |
| DPrivBench - Exponential Mechanism | 100 | Accuracy (0-1, times 100) on the report-noisy-max with Gumbe | 100 |
| DPrivBench - Mechanism Level | 99.5 | Accuracy (0-1, times 100) on the six mechanism tasks (mean o | 100 |
| DPrivBench - Permute-and-Flip | 99.6 | Accuracy (0-1, times 100) on the report-noisy-max with expon | 100 |
| DPrivBench - Report Noisy Max (Laplace) | 99.8 | Accuracy (0-1, times 100) on the report-noisy-max with Lapla | 100 |
| DiscoX | 76.66 | Overall Score | 100 |
| FlagEval VQA - Multi-Image Analysis | 62.5 | Score | 100 |
| FlagEval VQA - Visual Puzzles | 44.7 | Score | 100 |
| GAIA2 | 42.1 | Pass@1 (%) | 100 |
| GAIA2 - Ambiguity | 51.9 | Score (%) | 100 |
| GAIA2 - Execution | 69.2 | Score (%) | 100 |
| GAIA2 - Noise | 35.4 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-high · How It Works · Data refreshed daily, snapshot 2026-10-20.