PFMBench: Protein Foundation Model Benchmark

The first broad, standardized evaluation of protein foundation models: 17 models across 38 tasks spanning 8 areas of protein science, with a task-correlation analysis that maps which capabilities transfer. From Stan Z. Li's AI Lab (Zhangyang Gao, Cheng Tan, Stan Z. Li), it gives the protein-FM field a common yardstick analogous to what MMLU/GPQA provide for language models.