Large reasoning models often show counterintuitive behavior, putting more computational effort into simple tasks than difficult ones while producing worse results overall. Researchers have established ...