This feels like the early days of ISPs throttling VPN traffic. You're paying for a service with certain capabilities, then getting restricted for actually using those capabilities through a different interface.

The fundamental question is: if I'm a paying subscriber, why does it matter whether I access the model through your web UI or through an API wrapper? The compute cost is the same either way.

I suspect the real concern isn't usage volume but data pipeline control. When users interact through the native UI, Google gets structured interaction data. Through third-party tools, they lose that feedback loop.