GeneBench-Pro is a new benchmark that tests AI agents’ ability to navigate ambiguity and make consequential judgments in computational biology. It simulates real-world research scenarios, where data is messy and instructions are scarce. The benchmark evaluates AI models’ ability to make higher-order judgments, such as handling ambiguity and revising assumptions. GeneBench-Pro is designed to address the current bottleneck in biology research, where downstream computation and analysis are becoming the limiting factors. Why it matters: GeneBench-Pro has the potential to drive significant progress in AI-powered biology research, enabling scientists to focus on higher-level decision-making and accelerating breakthroughs in the field.