Drex Max reportedly edges Jev 1.14 in an exploratory AI-paper classification test

For DAIR.AI’s reading collections, the researcher compared Drex Max with Jev 1.14 using only paper titles and abstracts. They used 101 labeled papers to set decision thresholds, then tested the models on 149 papers. Across five repeated runs, Drex Max averaged 78.1% with both labels correct, versus Jev’s 74.9%. The author calls the work exploratory and says freshly labeled papers would provide a stronger test before relying on the decisions.