Results

Every published run, with its charts and its full per-task data. There is no composite score and no leaderboard. Each run reports raw numbers and states the conditions it was measured under.

No runs published yet.

Reading these numbers. A pass rate only means something alongside the conditions: which suite, which model, which harness, which Xcode, and, when a task was attempted more than once, which attempt counts. Every run page states all five.