“According to OpenAI's internal "GDP-Val" evaluation, the GPT-5.2 Pro model beats or ties human knowledge workers in 74.1% of tasks.”