GPT-6.2 Astra (Medium): benchmark results

Provider: OpenAI. Released 2026-11-17. Access: API.

Aggregate Rating 1986 ± 13, rank #12 of 2274 rated models, from 122 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
DeepsecBench-AA56.43Median F2 (%)100
AA Terminal-Bench Science 0.1 - Life Sciences70.18Pass@1 (%)99.7
DeepsecBench53.81Recall-weighted F2 score (%)98.3
AA GDP.pdf - Finance/Investing - Criterion Pass Rate92.29Criterion Pass Rate (%)98.2
AA Terminal-Bench Science 0.174.76Pass@1 (%)98.1
AA Terminal-Bench Science 0.1 - Earth Sciences79.17Pass@1 (%)98.1
AA Terminal-Bench Science 0.1 - Engineering Sciences74.07Pass@1 (%)98.1
AA GDP.pdf - Engineering - Criterion Pass Rate87.79Criterion Pass Rate (%)97.8
ARC-AGI-198.5Accuracy (%)97.4
ARC-AGI-293.33Accuracy (%)97.1
AA MMMU-Pro86.13Accuracy (%)96.8
AA FrontierCode 2.046.56Guardrail-gated Rubric Score (%)96.6

Interactive version: theaggregate.ai/model?slug=gpt-6-2-astra-medium · How It Works · Data refreshed daily, snapshot 2026-11-18.