FM SO.P: A Progressive Task Mixture Framework with Automatic Evaluation for Cross-Domain SOP Understanding

Researchers introduced FM SO.P, a new framework that teaches AI to understand Standard Operating Procedures (SOPs) across different industries. The system uses progressive task mixtures to build reasoning capabilities in stages, plus an automatic multi-agent evaluation system. Their 32B parameter model achieved a 48.3% pass rate on SOPBench, matching the performance of the much larger 72B Qwen-2.5 baseline while using significantly fewer computational resources.