GPT-5.5: benchmark results

OpenAI's standard GPT-5.5 model for general-purpose reasoning, coding, and writing. Provider: OpenAI. Released 2026-04-23. Access: API.

Unified ELO 1743 ± 1, rank #17 of 1392 rated models, from 863 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AGC-Bench - arastories1.28Dataset z-score100
AGC-Bench - brainteaser1.84Dataset z-score100
AGC-Bench - liveideabench1.57Dataset z-score100
AGC-Bench - newyorker_humor1.06Dataset z-score100
AGC-Bench - showerthoughts1.6Dataset z-score100
AI for Education Pedagogy92.1Accuracy (%)100
AI for Education Pedagogy - Primary96.71Accuracy (%)100
AI for Education Visual Maths - Statistics and Probability85.71Accuracy (%)100
ActiveVision10.6Accuracy (%)100
Age of LLM3Points per match (self-reported)100
AtmosCoder-Bench97.6Accuracy (%, mean of 3 runs)100
AutomataBench45.28Weighted pass@1 (%)100

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