Skip to content

Coding and terminal agents

Long-Horizon Terminal-Bench

46 tasks testing sustained useful work over hundreds of steps.

About Long-Horizon Terminal-Bench

What it measures

Hidden verifier success after rebuilding from the output artifact.

Scope & limitations

Long-running persistence test; do not confuse the LHTB publication with a numbered Terminal-Bench release.

Evaluation environment

Long-running terminal agent with stateful workspaces.

Source

Original benchmark

Dataset & task catalog ↗