OpenAI: GeneBench-Pro

- GeneBench-Pro is a new benchmark (a standardized test to compare AI models) released by OpenAI for evaluating AI systems that predict gene activity from DNA sequences. - This matters because understanding how DNA sequence leads to gene expression (which proteins a cell makes) is a central challenge in biology; better AI models could speed up drug discovery, disease understanding, and synthetic biology. - The benchmark tests models on 15 tasks across five categories: promoter activity (how strongly a gene is turned on), RNA splicing (how genes are cut and rejoined), expression levels, regulatory element effects, and variant effects (how a single-letter DNA change alters gene function). - OpenAI collected about 1,000 new experimental measurements specifically for this benchmark, aiming to address a common problem: existing AI models in genomics often seem accurate but fail on new, unseen data because they memorize patterns rather than learn real biology. - By releasing GeneBench-Pro publicly, OpenAI is inviting other researchers to test their own models on the same rigorous set of tasks, setting a clearer standard for progress in AI-driven genomics.