Index

Agentic tasks: an agent is given an instruction and a real environment, and is graded on the state it leaves behind rather than on what it says.

See: https://www.tbench.ai