OpenAI senior researcher Noam Brown says that if he had to choose between GPT-5.6 and a human research intern for most AI research tasks today, he'd go with the AI.
Brown made the comment during ICML 2026 when asked whether AI had reached the level of an AI research intern. He also acknowledged that there's currently no accepted way to measure whether AI truly operates at that level.
Another OpenAI researcher, Yash Pande, offered a more practical benchmark: how many GPUs researchers are willing to allocate to Codex for scientific tasks. If an AI is genuinely better than a human, the theory goes, companies would pour more compute into it.