HCAST: Human-Calibrated Autonomy Software Tasks

Benchmark of 189 machine-learning, cybersecurity, software engineering, and reasoning tasks with over 1,500 hours of human baselines, used to evaluate autonomous AI agents and to which EquiStamp-affiliated researchers contributed alongside METR.

Endorsements support EquiStamp.

No endorsements yet.

Discussion

Start Date

-

End Date

-

Expected Duration

-

Funding Raised to Date

-