GPT-5.4 Nano (xHigh): benchmark results

GPT-5.4 Nano evaluated at the xhigh reasoning-effort setting. Provider: OpenAI. Released 2026-03-17. Access: API.

Unified ELO 1628 ± 1, rank #325 of 1761 rated models, from 59 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA IFBench75.92Accuracy (%)94.5
AA Terminal-Bench Hard42.42Accuracy (%)89.2
AA CritPt9.25Accuracy (%)84.6
Artificial Analysis Intelligence Index30.92Intelligence Index82.1
LiveBench Table Join50.59Score81.1
AA Long Context Reasoning76.67Accuracy (%)80.8
AA Humanity's Last Exam28.27Accuracy (%)80
AA Omniscience - Software Engineering (SWE)38.1Accuracy (%)70.1
AA GPQA Diamond81.72Accuracy (%)69.7
AA Omniscience - Science, Engineering & Mathematics34.2Accuracy (%)68
AA TAU-2 Bench76.02Accuracy (%)67.8
AA-LCR72Accuracy (self-reported)67.1

Interactive version: theaggregate.ai/model?slug=gpt-5-4-nano-xhigh · How It Works · Data refreshed daily, snapshot 2026-09-05.