ModelsSep 8, 2026 · 06:355 sources

GPT-6 Astra Scores 14% on MazeBench, 7x Lead Over Claude Fable 5.1

From huggingnews.com

AI;DR

OpenAI's GPT-6 Astra scored 14% on the MazeBench benchmark, completing a 60-hour evaluation of long-horizon spatial reasoning without external Python code, outperforming Claude Fable 5.1's 2% score and GPT-5.6 Sol's 13% score.

Read the sourceSign in to suggest

Sources

+ 3 more
  1. SUPPORT@rohanpaul_ai“reduced a projected 3B-token trajectory to about 350M tokens”x.com
  2. SUPPORT@reach_vb“Astra is SoTA on MazeBench by a huge margin”x.com
  3. SUPPORT@andrewcurran_“massive jump with Astra”x.com
#gpt-6#maze-bench#spatial-reasoning#openai#competitive-benchmark#claudeMarkdown
Content history (2)
  • 2026-10-04 · Summary · vi · Replaced stored text
    Mô hình GPT-6 Astra vừa ghi nhận kết quả 14% trong bài benchmark MazeBench, cho thấy hiệu…
    Mô hình mới nhất của OpenAI đã hoàn thành đợt đánh giá kéo dài 60 giờ về khả năng suy luậ…
  • 2026-10-04 · Title · vi · Replaced stored text
    GPT-6 Astra đạt 14% trên MazeBench, bỏ xa Claude Fable 5.1 gấp 7 lần
    GPT-6 Astra đạt 14% trên MazeBench, dẫn trước Claude Fable 5.1 gấp 7 lần

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today