We evaluate AI models on the work attorneys actually do.
Percipient is a legal technology and services company building evaluation infrastructure, preference data, and RL/RLHF pipelines for frontier AI labs and other organizations building legal AI.
This is where we publish open artifacts from that work: legal-domain benchmarks and selected datasets that support reproducible measurement of model performance on real-world legal tasks. Check back. We will continue to update the benchmarks and add additional legal tasks.