Coasty Systems, Inc.

Teams building or evaluating computer-use AI agents need a way to know which models actually perform well on real-world tasks. Without a shared, transparent benchmark, claims about model performance are hard to verify or compare. The text indicates this is addressed by live tasks and blind human judgment rather than self-reported or opaque metrics.

Coarena by Coasty is a computer-use AI benchmark and model leaderboard that evaluates agents on live tasks using blind human judgment. It publishes model profiles, an evidence dataset, and a stated methodology, with every claim tied to a cited file. The product includes a CUA KnowledgeBench, a safety benchmark, defined metrics, an API, and a feature to watch two computer-use agents work on a research query submitted by the user.