GPT-5.4 (Low): benchmark results

GPT-5.4 evaluated at the low reasoning-effort setting. Provider: OpenAI. Released 2026-03-06. Access: API.

Unified ELO 1664 ± 1, rank #202 of 1761 rated models, from 44 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UGI - Natural Intelligence67.6NatInt Score97
UGI - Writing64.85Writing Score95.5
AA Omniscience - Software Engineering (SWE)79.9Accuracy (%)94.7
AA Omniscience - Health45.4Accuracy (%)93.3
AA Terminal-Bench Hard43.18Accuracy (%)90.3
AA-Omniscience Accuracy47.85Accuracy (%)90.3
AA Omniscience - Law39.3Accuracy (%)88.5
AA Omniscience - Humanities & Social Sciences42Accuracy (%)88
AA Omniscience - Business35.9Accuracy (%)87.8
IH-Benchmark94.1Overall Compliance (%)86.1
AA Omniscience4.78Score85.8
AA Omniscience - Science, Engineering & Mathematics44.6Accuracy (%)85.8

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