“On the Terminal Bench coding benchmark, Anthropic's Mythos and Fable-5 models scored 88%, ahead of GPT-5.5's score of 83.4%.”