“We found that GPT-4.1 reduces the number of times that it needs to read unnecessary files by 40% compared to other leading models, and also modifies unnecessary files 70% less,” Mohan said.