LisanBench creator Lisan al Gaib says Opus 4.8 beats GPT-5.5 among non-thinking models but ranks fifth overall · Digg