#Berkeley

1 articles

A digital representation of a robotic arm solving a complex logical puzzle, with glowing neural netw

GPT-5.5 Wins 'Agents' Last Exam' Benchmark in Surprise Upset

GPT-5.5 has won the new 'Agents' Last Exam' (ALE) benchmark developed by UC Berkeley, surpassing Claude Fable 5. The benchmark focuses on real-world professional task execution, marking a shift in industry competition toward AI agent capabilities.

Jason·Jun 11, 2026·2 min read

#Berkeley | Vela