Legacy ChatGPT models have historically stumbled on gastroenterology board exams. But DeepSeek, a new medical AI model, just changed the game—scoring a remarkable 81.5% when search-augmented, far outpacing ChatGPT-3 and ChatGPT-4, both of which failed to meet the passing threshold.
Tested on 455 questions from the American College of Gastroenterology, DeepSeek’s performance hints at a future where AI could supplement medical training—or even clinical reasoning.