Search bench primary metric

Count of OpenRouter search-benchmark runs for each stored primary metric. The label is the stored metric name. The number is that count. Bar length is the count against the highest count on this page. A longer bar is more runs. Each search bench is counted once, from the primary-score slice. This is not an accuracy score and not an item F1 score. A run with no stored primary metric stays off the page.

Loading the benchmark snapshot.