OpenDecider hits 0.796 on typed decisions

OpenDecider-nano, a roughly 400-million-parameter Apache-2.0 model, scored 0.796 on 2,000 typed-decision questions versus 0.766 for Laya and 0.754 for Jev in the project’s comparison. The Hugging Face report says OpenDecider-small matched Jev on 200 unseen general decisions (0.735 versus 0.730) and nano ran at 16 milliseconds per question on an L40S; the public repository notes Jev still leads on its application battery and the general set is small. Teams weighing Jev’s typed-decision economics can treat OpenDecider as a self-hostable baseline, then reproduce calibration and error costs on their own queue before granting it authority.