GPT-6 (Medium): benchmark results

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

Unified ELO 1778 ± 1, rank #13 of 1761 rated models, from 24 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
FrontierMath - Tier 4 (v2)97.56Accuracy (%, 41 private v2 problems)100
AA Omniscience - Software Engineering (SWE)90.2Accuracy (%)99
AA Omniscience42.22Score98.8
Artificial Analysis Intelligence Index52.25Intelligence Index98.6
AA GPQA Diamond93.94Accuracy (%)98.5
ARC-AGI-292.08Accuracy (%)98.4
AA Humanity's Last Exam52.73Accuracy (%)98.2
AA-Omniscience Accuracy60.57Accuracy (%)98.2
AA CritPt29.14Accuracy (%)98.1
AA Omniscience - Health52.4Accuracy (%)98
AA MMMU-Pro85.09Accuracy (%)97.7
AA Omniscience - Business50Accuracy (%)97.7

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