OpenAI sandbox task fails after 3 hours; Pro session grinds on for 5

teortaxesTex · x · 2026-09-08

User @teortaxesTex shared a hands-on test of an OpenAI sandbox environment: the task ran for about 3 hours before failing, which he suspects means he pushed the sandbox over the edge. As an interesting comparison, a Pro session without Deep Research had been running for almost 5 hours with barely any visible progress — a result he called "bullish". He's also waiting to see what the accompanying deep research session yields.

Related event: Users Test Marathon AI Agent Tasks on Gemini and OpenAI With Little Progress(2 posts)→

Original post →

More from Fun

Fun channel →