Admittedly, I did always feel like being able to use the assistant up to the cap of your monthly/yearly subscription price was "too good of a deal" even if you had built in a margin into the token price.
I understand having to cut back the included AI budget in subscriptions, but offering a one time "trial" budget, seems like you're removing a feature, that some people may have subscribed to for in the first place.
I use some cheap models like kimi k2 fairly regularly, but less than a dollar a month. And I can imagine many people are in the same boat, not using LLMs heavily, but it's a nice to have when you need it. Knowing that the budget replenishes, and that typically im far from the limit makes it a nice feature included in the bundle rather than needing to seek an LLM elsewhere.
If you must scale back how generous you are with AI, I think it would be much more fair to just limit how much budget is included each month/year for LLM use. It's a much easier pill to swallow, as it won't feel like loosing a feature entirely.
Also, fwiw, I noticed quick getting worse in recent times, is that cause you switch to a cheaper model? If so, that's a shame, and quite honestly I'd rather have the usage of a slightly more expensive model even if it means charging the token cost within the same budget.
