According to CoinMeta, OpenAI has released a computational biology evaluation benchmark Genebench-Pro to test the multi-step decision-making capabilities of AI agents in complex scientific research scenarios such as genomics and translational medicine. The new benchmark consists of 129 questions (82 of which have been reviewed by external experts), and it generates data with clear causal relationships through computer simulations to prevent models from cheating by taking shortcuts or catering to the preferences of the question setters. Test results show that even top models struggle significantly when dealing with scientific reasoning that involves quantitative uncertainties, with the pass rate for GPT-5.6 being only 31.5%. In addition, former researcher Andrew Ho of OpenAI has left to start his own business, focusing on creating high-quality reinforcement learning data for large models. It is expected that over $100 billion will be invested in precise training data in the future.