Scores are desk assessments of published materials; security scans are shown separately, and no score here uses production telemetry. Each row carries the evidence class behind its number.
Last update: 19 October 2026. Scores are re-assessed every week against each agent's public materials; changes of more than 5 points are checked by a person before they publish. We are not adding new agents right now.
New = Added in the last 30 days.
Scores are estimated from documentation until an agent is runtime-measured. Methodology v2.6.0 rev 1.27.
Try clearing the active filter, or pick a different traction bucket. The full leaderboard is one click away.