OpenAI GeneBench-Pro: AI Scores 32% on Biology Intuition

OpenAI has unveiled a new computational biology benchmark called GeneBench-Pro. This benchmark measures the scientific research capabilities of artificial intelligence. What's interesting is that the top AI model, GPT-5.6 Sol Pro, scored just 31.5% on these challenging problems. Other models, like Anthropic's Claude Opus 4.8, scored even lower at 16%. GeneBench-Pro features 129 synthetic problems covering genomics and other biology areas. It aims to evaluate "research intuition," which is the ability to make high-level judgments with data, not just recall facts. OpenAI estimates a human expert would take 20 to 40 hours to solve a single problem. While current AI isn't reliable enough to replace human experts, even partial automation could offer significant value. The bottom line is that even the best AI still gets about 70% of these real biology problems wrong, showing the current limits of the technology.