GPT-5.6 — benchmark results
Provider: OpenAI. Access: API.
Unified ELO 1980 ± 33, rank #27 of 1723 rated models, from 91 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA CritPt | 32.29 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - HTML | 92 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - PHP | 96 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 92.22 | Accuracy (%) | 100 |
| AA Terminal-Bench Hard | 65.91 | Accuracy (%) | 100 |
| ARC-AGI-2 | 92.5 | Accuracy (%) | 100 |
| AutomationBench | 18.1 | Pass Rate (%) | 100 |
| FrontierMath - Tiers 1-3 (v2) | 89.12 | Accuracy (%, 285 private v2 problems) | 100 |
| Gert Labs Rankings | 73.77 | GScore (%) | 100 |
| KernelBench Hub - Hard | 30 | Best % of Hardware Roofline | 100 |
| Tau3 Banking | 32.99 | Success Rate (%) | 100 |
| Vals AI CyberBench | 88.14 | Accuracy (%) | 100 |
Interactive version: aibenchmarks.dev/model?slug=gpt-5-6 · How the rankings work · Data refreshed daily, snapshot 2026-07-10.