Real-person blind evaluation list for computer-operated intelligent agents
Coarena is an evaluation platform launched by YC-accelerated company Coasty. It is specifically designed to compare intelligent agents that are capable of using computers and browsers in the same way humans do. The platform operates on a competitive format: you submit a real-world task, such as conducting research or planning a journey, and the system allows two anonymous intelligent agents to complete it simultaneously. After observing their performance, you can vote to determine which agent performed better.
Based on these blind evaluation results, the ranking system will assign scores to the models according to several criteria such as human preferences, task completion efficiency, speed, ability to recover after errors, and cost. The website also offers benchmark tests, model profiles, security assessments, public datasets, methodological documentation, as well as a Metrics API, and ensures that every numerical value can be traced back to the specific rules and documents that generated it.
The platform is available for free. After logging in, users can assign tasks and participate in voting. It is suitable for developers and enterprises that are in the process of selecting intelligent agents for computer operations, as well as for researchers interested in the progress in this field.