Tag
Introduces Agents' Last Exam (ALE), a benchmark for evaluating AI agents on long-horizon, economically valuable real-world tasks across 13 industry clusters with over 1000 tasks, revealing a large gap between benchmark performance and practical deployment.