The shift in licensing specifically targets "hyperscalers"—large-scale cloud service providers—and represents a departure from the open-source standards used by many competitors.
By moving away from unrestricted access for the world's largest firms, Z.ai may be attempting to better monetize its most advanced technology or maintain tighter control over the inference layer—the specialized environment where AI models process data to generate responses.
The source indicates that as Chinese open-weight models begin to match the performance of leading U.S.
labs, they are increasingly viewed as valuable commercial assets rather than just community contributions.
Technically, GLM-5.3 is a mixture-of-experts architecture featuring 753 billion parameters and a massive 1 million-token context window, which allows it to process vast amounts of information in a single request.
Despite being "open weight," the model’s massive size makes it difficult for average users to run locally, as even compressed versions require roughly 245GB of memory.
While individual developers and smaller firms can still use the weights freely under the new terms, the largest global tech entities now face significant administrative hurdles to integrate the model into their own cloud computing infrastructures.