GPT-5: benchmark results
OpenAI's standard GPT-5 model for broad reasoning, coding, writing, and tool-use tasks. Provider: OpenAI. Released 2025-08-07. Access: API.
Unified ELO 1690 ± 1, rank #66 of 1392 rated models, from 1155 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ASCIIEval | 55.9 | Macro-Avg Accuracy (%) | 100 |
| AfroBench | 77.74 | Average Score (%) | 100 |
| BIRD-Interact (a-Interact) | 25.52 | Normalized Reward | 100 |
| Beyond Binary | 84.52 | NDCG@All (self-reported) | 100 |
| Blueprint-Bench | 0.43 | Similarity Score | 100 |
| Bongard in Wonderland | 64 | Problems Solved (out of 100) | 100 |
| BrowserGym WorkArena L2 | 69.4 | Success Rate (%) | 100 |
| CAIA - Pass@1 (With Tools) | 70.2 | Pass@1 (%) | 100 |
| CAIA - Pass@1 (Without Tools) | 28.1 | Pass@1 (%) | 100 |
| CMT-Benchmark | 30 | Pass@1 (%) | 100 |
| Can Large Language Models Handle Discourse Par | 72.4 | Overall Avg (MS) (self-reported) | 100 |
| CodeClash - Poker | 1599 | ELO | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5 · How It Works · Data refreshed daily, snapshot 2026-09-05.