GPT-5.4 (xHigh) — benchmark results
GPT-5.4 evaluated at the xhigh reasoning-effort setting. Provider: OpenAI. Released 2026-03-06. Access: API.
Unified ELO 1997 ± 21, rank #25 of 1776 rated models, from 122 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CLBench | 27.9 | Solving Rate (%) | 100 |
| MathArena - ArXiv Math Jan 2026 | 76.09 | Accuracy (%) | 100 |
| PlanningBench | 63.17 | All-pass (%) (self-reported) | 100 |
| Sandboxed Coding Agents are Competitive Omni-m | 75 | OmniGAIA Avg. (self-reported) | 100 |
| AA Omniscience - Software Engineering (SWE) - Python | 90 | Accuracy (%) | 99.6 |
| AA Omniscience - Software Engineering (SWE) - Julia | 84 | Accuracy (%) | 99.4 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 91.11 | Accuracy (%) | 99.4 |
| AA Omniscience - Software Engineering (SWE) - Java | 71 | Accuracy (%) | 99.2 |
| ZeroBench | 23 | Score (%) | 99.2 |
| AA SciCode | 56.6 | Accuracy (%) | 99.1 |
| UGI - Natural Intelligence | 73.53 | NatInt Score | 99.1 |
| AA Omniscience - Software Engineering (SWE) - JavaScript | 88.18 | Accuracy (%) | 99 |
Interactive version: theaggregate.ai/model?slug=gpt-5-4-xhigh · How the rankings work · Data refreshed daily, snapshot 2026-07-22.