Run independently by Artificial Analysis at the highest reasoning effort. Elo is a pairwise-judging rating that moves with each evaluation round; scores from different snapshots are not overwritten.Artificial Analysis's long-horizon knowledge-work benchmark: models work through multi-week projects with thousands of linked input files; grading combines rubric scoring and pairwise judging by a frontier-model panel. This criterion records the rubric pass rate, kept separate from the Elo rating.