GLM 5.3 is Z.ai's August 2026 open flagship—same 743B GLM-5.2 base, with all gains from scaled post-training for coding agents, long-horizon work, and tool follow-through. Our assistant tests put it ahead of 5.2 on helpfulness, instruction following, and comprehension for messy multi-step help that has to finish. Empathy stays cool and professional. Creative writing is still utilitarian: clearer diffs than stories, so Fable and GPT-4.5 remain the prose picks. Best open model when you want an agent that keeps going; not the warmest chat partner. Same $1.40/$4.40 API rate as 5.2.