“The GPT 5.3 Codex model is significantly more efficient than previous versions.”