I find it astounding that ppl still comment “it’s still behind” or “it’s not the best model”. Everything is about the harness. Even the big AI labs are focusing on managing agents - sandboxes, memory, context, skills, loops. With the right harness GLM 5.2 can do no wrong.