“According to OpenAI's internal "GDP-Val" evaluation, the GPT-5.2 "thinking" model beats or ties human knowledge workers in 70.9% of tasks.”