Sefer beats frontier AI models

We put Sefer head-to-head with the top frontier models on real HTS classifications from CBP's CROSS rulings. It comes out ahead on heading and full-code accuracy, matching-digit depth, and hallucination rate. See the latest numbers below.