Is there one leaderboard for agent benchmarks?

No. Each agent benchmark runs its own leaderboard, scored against its own tasks, and nobody has merged them into a single ranking. BFCL’s leaderboard only ranks tool calling, SWE-bench Pro’s only ranks coding, and GAIA2’s only ranks the tasks GAIA2 sets; a model’s place on one says nothing about where it lands on another. The closest thing to a combined view, Princeton’s Holistic Agent Leaderboard, still keeps nine benchmarks as nine separate tables rather than one score. HAL’s own site gives the reason: “Agents can be 100x more expensive while only being 1% better,” and “Downstream developers can’t tell the difference from 1D leaderboards.” HAL has since paused adding new models to focus on measuring reliability instead of ranking. Picking a benchmark means picking the task family you care about first; there’s no index that picks it for you.