Token based is a pretty strong downside for me that would be enough to get me to use other tools like Cursor (even though I love JetBrains IDEs). I get actively stressed watching an automated system burn through my money on its own recognizance. If I'm going to have quotas or usage-based pricing I want the metrics used to be things that I have direct control over, not things that the service provider controls.
TANSTAAFL. With flat pricing, companies have incentives to downgrade you to cheaper models - which currently strongly correlates with worse quality of output - or, more likely, to trim context significantly and hope you won't notice.
But yes, there should absolutely be ways to track usage, ideally before the prompt is even submitted for processing (maybe for >N tokens per query, where N can be specified in settings but has a reasonable default).
