Agent's Last Exam
An AI evaluation benchmark co-led by UC Berkeley, designed with the participation of 300 experts across 55 industries. It employs a fully automated scoring mechanism, focuses on resistance to training data contamination, and covers highly specialized domains such as legal compliance, bioinformatics, and financial modeling.
Timeline (last 90 days)
DeepSeek V4.1 Flash 1017 rose from 31.2 to 42.3 on Terminal Bench 4.0, from 20.2 to 42.3 on Program Bench, from 31.8 to 35.4 on Agent's Last Exam, from 64.6 to 72.1 on Apex Bench, from 78.9 to 83.0 on Chartography, and from 89.6 to 92.5 on Baby Vision.
GPT-6 Astra set a new SOTA on Agent's Last Exam, designed by UC Berkeley, at a cost far lower than Fable 5.
In three benchmarks—Agent's Last Exam, ZeroBench, and DeepSWE—V4-Flash-Vision-Exp outperformed Opus 4.8.
All Facts (2)
DeepSeek V4.1 Flash 1017 rose from 31.2 to 42.3 on Terminal Bench 4.0, from 20.2 to 42.3 on Program Bench, from 31.8 to 35.4 on Agent's Last Exam, from 64.6 to 72.1 on Apex Bench, from 78.9 to 83.0 on Chartography, and from 89.6 to 92.5 on Baby Vision.
50%UnverifiedGPT-6 Astra set a new SOTA on Agent's Last Exam, designed by UC Berkeley, at a cost far lower than Fable 5.
50%