“OpenAI developed the HealthBench benchmark, which uses 5,000 conversations and 49,000 different criteria to evaluate the performance of large language models in healthcare.”