🧬 OpenAI Introduces GeneBench-Pro Benchmark for Biological Research
OpenAI has developed GeneBench-Pro — a benchmark for evaluating the ability of AI agents to perform scientific reasoning in biology and genetics. The test includes 129 tasks across 10 domains, requiring models to do more than simple fact-finding; they must make complex research decisions: data cleaning, selecting statistical methods, and adjusting plans upon discovering artifacts. The leader of the test was the GPT-5.6 Sol model.
🌍 GeneBench-Pro shifts AI evaluation from simple instruction following to testing "research taste" — the ability to handle uncertainty and data errors, which is critical for automating biomedical research.
👤 This is a step toward creating autonomous AI laboratories capable of saving thousands of dollars and dozens of hours of expert work by performing complex analysis of biological data with minimal human involvement.
Source 1: https://openai.com/index/introducing-genebench-pro/ Source 2: https://cdn.openai.com/pdf/21938268-21af-442f-af93-3b2249afb241/genebench-pro.pdf