#基準測試

2 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

A graphical visualization of sound waves changing from sharp, synthetic jagged lines to organic, smo

Tech Frontline

Scale AI Launches 'Voice Showdown' to Benchmark Real-World Voice AI Performance

Scale AI has launched 'Voice Showdown,' a benchmark testing voice AI against real-world human interactions instead of synthetic data, pushing the industry toward more resilient, practical applications.

Jason·Mar 22, 2026·2 min read