Nace AI releases Drex open decision models, claims lead over Jev and Claude Opus 5.5

Nace AI releases Drex, four open decision models for probability-based choices, in 8B, 9B, 35B (3B active) and 27B sizes.

Nace AI says its 27B Max model scores 67.52 on the public Decision Index 0.4.1 suite, against 60.70 for Jev 1.14.0, and answers in 183 ms, against 3.5 seconds for Claude Opus 5.5. It also says reinforcement learning from adjusted feedback cut the rate of answers that flip when option order changes from 6.3% to 4.2%, and that the 27B model lands within 1.6 points of Claude Opus 5.5 on Decision Bench v4. These are the company's own benchmark results.

A follow-up from Dongxi NLP, citing the Nace AI Drex Technical Report, describes the training as supervised fine-tuning, RLAF with KL regularization, and vision fine-tuning, and reports 94.0% accuracy for Drex Max on Decision Bench v4. Nace AI's post says Drex Max 1.1 and Drex for Finance are closed source and aimed at enterprise use, so the open release covers the four model family only.