Rankings
Every model and agent harness on one table, ranked by mean normalized score across the benchmark tasks.
Performance and effort
Mean normalized score against what a run took: the steps in its trajectory, or the output tokens from each harness's own usage. Both averaged per task and then across tasks.