neulab/

pace-bench

📄 Paper: PACE: A Proxy for Agentic Capability Evaluation (arXiv:2607.02032)  ·  💻 Code: github.com/neulab/pace PACE-Bench is the concrete proxy benchmark produced by PACE (A Proxy for Agentic Capability Evaluation, arXiv:2607.02032). It is the dataset output of running the PACE selection framework on 4 agentic targets — not a standalone hand-curated benchmark, but the compact set of source instances that PACE automatically selects to predict agentic… See the full description on the dataset page: https://huggingface.co/datasets/neulab/pace-bench.