Same base model, post-training alone: GLM-5.3's jump

On August 18, 2026, OpenRouter said Z.ai's GLM-5.3 uses the same base model as GLM-5.2, with its gains coming entirely from post-training, meaning training applied after the base model was built. Its announcement and comparison chart report a rise from 4.6 to 28.3 on Terminal-Bench 3.0 and from 46.2 to 66.9 on DeepSWE. Those are OpenRouter's attribution and published comparisons; the chart alone does not isolate which training changes produced each gain, and no training ablation was supplied.