Surprise upset: GPT-5.5 beats Claude Fable 5 on brutal new Agents Last Exam benchmark | Inshort Stories