Proprietary openai model, git patch, and sonet

if claude-3-sonet can calculate git patch’s better than the gpt4 proprietary model, why are we still using the proprietary model for auto edits?