Type: url
Original: https://www.tbench.ai/
Status: generated
Summary
Terminal-Bench is a benchmark, at version 4.0 at the time of this fetch, that its own site describes as measuring and evolving with the frontier of agent work. The landing page publishes a leaderboard ranking model and agent pairings by resolution rate alongside cost and token figures, with whiskers spanning a 95% confidence interval, and points to a task list hosted on the Harbor framework’s dataset hub. It names Stanford, Harbor and the Laude Institute as its hosts.