A test for how well an AI Agent completes tasks with several steps.
It is a driving test for AI Agents. Knowing road signs is easy. Now park without hitting the cones.
It gives an Agent a real-style task. It checks if the Agent can use tools and finish each step.
Agent
An Agentic Benchmark tests an Agent's planning, tool use, and task skills.
Evaluation harness
The Evaluation harness runs the tasks and records the results.
Leaderboard
Benchmark scores help a Leaderboard compare different Agents.