[AINews] Death of Params: Z.ai CEO Jie Tang on GLM 5.3 and the new Post-training Scaling Law
GLM-5.3 proves post-training RL on long-horizon tasks beats parameter scaling for reasoning
“Parameter count is only meaningful alongside three others — how much data you have, where you intend to spend your compute, and who will run the model, under what conditions.”
Z.ai's GLM-5.3 achieves its capability gains entirely through RL training on long-horizon synthetic environments — not parameter increases — with tasks designed to represent days of real engineering work. The team built fully synthetic pipelines to generate and verify training environments at scale, advancing recursive self-improvement. This challenges the industry's parameter-count shorthand and signals that post-training environment quality is becoming the new scaling frontier.