“On the HealthBench benchmark, Anthropic's Mythos-5 and Fable-5 scored 66%, compared to GPT-5.5's 51.8%.”