Coarena

Compare computer-use AI models through blind human evaluations and live benchmark tasks

Pricing model

Free

Rating

★ People love it

Time to value

Immediate

You can use Coarena to compare and benchmark computer-use autonomous agents against real-world digital tasks. The service lets you submit prompt challenges such as web research or spreadsheet editing, watch two competing models execute steps side by side, and rate outcomes through blind human testing. You can explore public leaderboards scored across task completion, operational speed, error recovery, and compute costs. In addition, you can access safety evaluations, inspect detailed model profiles, review reproducible benchmark rules, and analyze how agents operate desktop applications.

What Coarena does

Submit real-world computer tasks for agent head-to-head battlesVote on agent performance in blind evaluation roundsFilter leaderboards by completion rate, speed, and costReview reproducible methodology rules for model rankingsInspect model profiles and individual evaluation tracesQuery evaluation data through the metrics APISide-by-side agent battles on live tasksBlind human judgment for evaluating outputsLeaderboards tracking completion rates and recoveryCost and latency tracking per taskDedicated CUA KnowledgeBench benchmarksSafety and risk evaluation metricsAPI access to historical model performance metrics