Z.ai documents GLM-5.3 as a post-training-only upgrade over GLM-5.2 and reports 84.5 percent on CyberGym, with the API not yet available

Sources: primary, discussion
Summary: Z.ai's developer documentation describes GLM-5.3 as a post-training-only upgrade over GLM-5.2, with no change to pretraining. The page reports 84.5 percent on CyberGym. Every figure on the page is vendor-reported with no independent reproduction, and the same page states the API is not yet available.
Why it matters: An open-weights model reporting the top CyberGym score moves vulnerability-discovery capability out of the closed frontier labs, and the API to use it has not shipped.