GPT-6.2 Astra Pro: benchmark results

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

Aggregate Rating 1990 ± 11, rank #2 of 10489 rated models, from 14 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
DuelLab GameBench 2 - Reasoning Variants97.15GameBench2 normalized score (0–100)99.7
Arabic Broad Leaderboard - Translation (incl Dialects)8.19Average Score (0-10)99.2
Arabic Broad Leaderboard - MMLU9.92Average Score (0-10)98
MineBench2134Elo Rating97.6
Arabic Broad Leaderboard9.14Average Score (0-10)93
Arabic Broad Leaderboard - Trust & Safety9.67Average Score (0-10)91.8
Arabic Broad Leaderboard - Reasoning & Math9.33Average Score (0-10)89.1
Arabic Broad Leaderboard - RAG QA9Average Score (0-10)88.3
SimpleBench79.2Score (AVG@5)86.3
Tinybird AI SQL Benchmark - Exactness53.91Result exactness vs human reference queries (0-100)79.6
Arabic Broad Leaderboard - General Knowledge8.65Average Score (0-10)55.5
Tinybird AI SQL Benchmark - First-Attempt Success Rate94Questions answered with a valid query on the first attempt (52.9

Interactive version: theaggregate.ai/model?slug=gpt-6-2-astra-pro · How It Works · Data refreshed daily, snapshot 2026-11-20.