GPT-6 Astra Pro: benchmark results

OpenAI GPT-6 Astra in Pro reasoning mode. Kept separate from standard Astra in benchmark results; launch announced September 3, 2026 with staged availability. Provider: OpenAI. Released 2026-09-03. Access: API.

Aggregate Rating 1965 ± 7, rank #4 of 10556 rated models, from 23 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
MineBench2206Elo Rating100
WeirdML93.57Average Score100
Arabic Broad Leaderboard - Translation (incl Dialects)8.17Average Score (0-10)98.8
Chess Puzzles (Epoch AI)76Accuracy (%)98.8
OpenRouter GPQA Diamond95.5Accuracy (%)98.7
Arabic Broad Leaderboard - MMLU9.92Average Score (0-10)98.4
Arabic Broad Leaderboard - Trust & Safety10Average Score (0-10)98
Last Translation Benchmark45.33Verified Translations (%)98
Arabic Broad Leaderboard9.2Average Score (0-10)96.1
SimpleBench86.5Score (AVG@5)95.5
The Aggregate - Trip Outcome Prediction (5 Questions)69.8Accuracy (%)95.2
Arabic Broad Leaderboard - Writing (incl Dialects)7.96Average Score (0-10)93.7

Interactive version: theaggregate.ai/model?slug=gpt-6-astra-pro · How It Works · Data refreshed daily, snapshot 2026-11-17.