Develop novel, difficult evaluation tasks for frontier AI models; verify task specifications and solvability; baseline and score model and human completions; and improve task development infrastructure and workflows to support METR's Time Horizons evaluations.