Towards Foundation Decision Models: Drex

Researchers have introduced Drex, a family of four open decision models with 8B, 9B, 35B, and 27B parameters designed to return probability distributions over user-supplied options without autoregressive decoding. The models are trained using supervised fine-tuning, reinforcement learning from adjusted feedback, and vision tuning for the Max variant, with ablation studies showing improvements in accuracy and robustness.