GPT-6 (High): benchmark results

Provider: OpenAI. Released 2026-09-03. Access: API.

Unified ELO 1797 ± 1, rank #7 of 1761 rated models, from 24 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA Omniscience43.73Score100
AA MMMU-Pro86.42Accuracy (%)99.6
AA GPQA Diamond94.95Accuracy (%)99.4
ARC-AGI-198.5Accuracy (%)99.3
Artificial Analysis Intelligence Index53.36Intelligence Index99
AA Omniscience - Health53.7Accuracy (%)98.9
AA Omniscience - Software Engineering (SWE)89.9Accuracy (%)98.8
AA-Omniscience Accuracy61.13Accuracy (%)98.6
AA Humanity's Last Exam53.06Accuracy (%)98.5
ARC-AGI-292.08Accuracy (%)98.4
AA Omniscience - Humanities & Social Sciences58.1Accuracy (%)98.2
AA Omniscience - Business50.5Accuracy (%)98.1

Interactive version: theaggregate.ai/model?slug=gpt-6-high · How It Works · Data refreshed daily, snapshot 2026-09-05.