Open benchmark

The agentic AI leaderboard where the receipts are public.

Every score below came from an agent driving a real multi-step task against a live tool server, recorded call by call and graded on a published rubric. Click any row and read the entire run.

Take the benchmark How scoring works

Runs graded

38/100

Average score

Pass rate

No published runs yet. The board opens with the first one.

We will not seed this with invented names or padded scores. The benchmark is live, the rubric is published, and the first engineer to run it and publish gets the top of an empty board.

Be the first Read the rubric first

Runs appear here only when the engineer chooses to publish them. Scores are never edited after grading, and a published run keeps its full trajectory attached so any employer can audit it.