OpenAI Releases GeneBench-Pro to Measure AI Performance in Computational Biology

My take: This one got my curiosity... and then made me raise an eyebrow. OpenAI just dropped GeneBench-Pro, a new evaluation benchmark that measures how well AI agents handle complex computational biology tasks: genomics, translational medicine, the kind of serious science that matters. 129 problems with deliberately noisy data to simulate real-world conditions. Top score goes to GPT-5.6 Sol Pro at 31.5%, with Claude Opus 4.8 reaching 16%. So the one who writes the test also walks away with the top grade. It is like if your chemistry teacher designed the exam and then showed up in class with their diploma in hand 🙄. What did impress me: at just 31.5%, the best model in the world still fails 7 out of 10 real biology problems. Not a knock, just proof of how genuinely hard real science is. For those of us building AI in health or research, this benchmark exists and it measures something real.