Skip to content

Scientific and ML agents

LAB-Bench (FutureHouse)

Biology literature, database, figure and procedural-reasoning questions.

About LAB-Bench (FutureHouse)

What it measures

Subtask answer accuracy.

Scope & limitations

Different from Harvey LAB. The Harbor 181-task port is not the full upstream multi-subtask corpus.

Evaluation environment

Science assistant with track-appropriate tools.

Source

Original benchmark

Dataset & task catalog ↗