MedXpertQA MM

Anthropic's Claude Fable 5 model demonstrates high accuracy on biomedical benchmarks but refuses to answer 8% to 99.4% of questions depending on the test, revealing a reluctance to engage with certain…