Coarena

Coarena is an evaluation platform for computer-use AI agents, hosting live tasks where agents attempt real actions while human reviewers provide blind judgment on the results. The platform emphasizes transparent methodology, publishing the specific rule or metric behind every number it reports rather than presenting scores without context. A public leaderboard tracks agent performance over time, alongside dataset access for those who want to dig into the underlying task data. It's built for developers and researchers working on computer-use agents who need independent, real-world performance signals beyond a lab's own benchmark claims. The site is backed by Y Combinator and run by Coasty Systems, Inc., and includes a live demo where visitors can watch agents attempt tasks and see battles awaiting human evaluation. No pricing is listed on the site, and access to the benchmark and its API appears to be open rather than gated behind a signup or waitlist. As a very new platform, its leaderboard currently reflects a limited set of agents and tasks, so its rankings will likely shift significantly as more agents and evaluators join.