GPT-6 (Max): benchmark results

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

Unified ELO 1805 ± 1, rank #4 of 1761 rated models, from 68 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA MMMU-Pro86.88Accuracy (%)100
ARC-AGI-295Accuracy (%)100
AutomationBench41.4Pass Rate (%)100
Chess Puzzles (Epoch AI)72Accuracy (%)100
Epoch AI - Ebr Bench76.19Score100
Epoch AI - Frontiermath Erdos2.94Score100
Epoch AI - Mystery Game Puzzles84Score100
FrontierMath - Tiers 1-3 (v2)93.68Accuracy (%, 285 private v2 problems)100
LiveBench Plot Unscrambling86.29Score100
LiveBench Table Join58.37Score100
SimpleQA Verified75.6Accuracy (%)100
Vals AI Code Migration67.74Accuracy (%)100

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