Yesterday, an interesting paper was released by Prescient Design (Genentech), "Evaluating the Progression of Large Language Model Capabilities for Small-Molecule Drug Design." The authors formulate a variety of problems in small-molecule drug design as reinforcement-learning (RL) problems and use RL-based post-training to improve the performance of a 30B-parameter Qwen model (specifically, Qwen3-30B-A3B-Thinking-2507). They then compare the