To understand and predict the societal impacts of highly autonomous AI systems, we need benchmarks with grounding, i.e., metrics that directly connect AI performance to real-world effects we care about. We present HCAST (Human-Calibrated Autonomy Software Tasks), a benchmark of 189 machine learning engineering, cybersecurity, software engineering, and general reasoning tasks. We collect 563 human baselines (totaling over 1500 hours) from people skilled in these domains, working under identical conditions as AI agents, which lets us estimate that HCAST tasks take humans between one minute and 8+ hours. Measuring the time tasks take for humans provides an intuitive metric for evaluating AI capabilities, helping answer the question "can an agent be trusted to complete a task that would take a human X hours?" We evaluate the success rates of AI agents built on frontier foundation models, and we find that current agents succeed 70-80% of the time on tasks that take humans less than one hour, and less than 20% of the time on tasks that take humans more than 4 hours.
@article{arxiv.2503.17354,
title = {HCAST: Human-Calibrated Autonomy Software Tasks},
author = {David Rein and Joel Becker and Amy Deng and Seraphina Nix and Chris Canal and Daniel O'Connel and Pip Arnott and Ryan Bloom and Thomas Broadley and Katharyn Garcia and Brian Goodrich and Max Hasin and Sami Jawhar and Megan Kinniment and Thomas Kwa and Aron Lajko and Nate Rush and Lucas Jun Koba Sato and Sydney Von Arx and Ben West and Lawrence Chan and Elizabeth Barnes},
journal= {arXiv preprint arXiv:2503.17354},
year = {2025}
}