Open Agent Leaderboard Launched for General-Purpose AI Agents

An open evaluation framework called the Open Agent Leaderboard has been launched to benchmark the performance of full agent systems, not just the underlying models. The leaderboard reports both the quality and cost of agents across six benchmarks, testing diverse tasks like coding, customer service, and personal assistance. It aims to provide transparency on how well general-purpose AI agents perform in realistic scenarios, noting that model choice remains the dominant factor in performance.