GPT-5 (Medium): benchmark results
GPT-5 evaluated at the medium reasoning-effort setting. Provider: OpenAI. Released 2025-08-07. Access: API.
Unified ELO 1651 ± 1, rank #243 of 1761 rated models, from 163 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HAL Online Mind2Web | 42.33 | Accuracy (%) | 100 |
| HAL USACO | 69.06 | Accuracy (%) | 100 |
| Step Game (Lechmazur) | 5.49 | TrueSkill μ | 100 |
| SudokuBench (Single-shot, 4x4) | 86.7 | 4x4 Solve Rate (%) | 100 |
| Translation (Lechmazur) | 8.69 | Mean Score | 100 |
| AI for Education Pedagogy - Social studies | 90.91 | Accuracy (%) | 98.9 |
| Aider polyglot coding leaderboard | 86.7 | Pass rate (%) | 98.5 |
| Elimination Game (Lechmazur) | 5.97 | TrueSkill μ | 98.3 |
| MATH Level 5 | 97.92 | Accuracy (%) | 98.1 |
| AA Omniscience - Software Engineering (SWE) - PHP | 58 | Accuracy (%) | 96.8 |
| AA Omniscience - Software Engineering (SWE) - Dart | 42 | Accuracy (%) | 96.4 |
| MCP-Universe | 43.72 | Overall Success Rate (self-reported) | 96.3 |
Interactive version: theaggregate.ai/model?slug=gpt-5-medium · How It Works · Data refreshed daily, snapshot 2026-09-05.