GPT-6 Astra benchmark roundup: 99.9% on ARC-AGI-3, 100% on ExploitBench

CyrilXBT @cyrilXBTOriginally posted 2026.09.04

Open original on X

This is gpt-6 astra, and it might be the model that ends the chatbot era. anything you can do on a computer, it can do for you, fast. 99.9% on arc-agi-3 with a full harness. 98% on frontiermath tier 4. 100% on exploitbench, openai's toughest cybersecurity benchmark, beating gpt-5.6 sol's 78.5%. on real computer use, 72.6% on oswork 2.0 in roughly 40 minutes, versus sol's 65.7% in 75, nearly half the time for a better result. greg brockman's own words: "welcome to the agi era."

This post quotes another post: “GPT-6 Astra announced with a claim to do anything on a computer”