Apple now sells $49.99 AI top-ups. How do you price AI features that cost you per use?

Even Apple has stopped pretending AI features are free to run. I'd like to hear how other small teams are handling the meter, because I don't think anyone has this solved.

On 9 November Apple started selling what it calls intelligence usage packs. Per 9to5Mac and iClarified's write-ups, it's a $49.99 one-time purchase, sold first inside Logic Pro, for people on a Creator Studio subscription who have hit their monthly limit on the AI features. Apple's own terms page for the packs says they don't renew, expire 12 months after purchase, are non-refundable except where the law requires, and, in the line I found most interesting, that Apple "does not guarantee that a specific amount of the Usage Packs will result in the same output." The reports say Siri and general Apple Intelligence features aren't covered, and that the monthly allowance itself isn't stated in numbers.

I'm not here to judge whether $49.99 is fair. What caught my attention is that the most polished consumer company in the world landed on the same awkward compromise a lot of us have: a subscription with a soft ceiling, plus an extra charge for people who go past it, measured in a unit the customer can't really predict.

If you build anything with a model behind it, you know why. Every generation costs you something. A flat subscription works until a small group of heavy users eats the margin from everyone else. Pure usage pricing is honest, but it puts a taxi meter in the corner of the screen, and people use things differently when they can see it running.

The options I keep seeing, roughly:

A flat price with a fair-use cap that most people never hit. Simple to explain, and the cap is a quiet insurance policy.

Credits. Clear for the business, but customers often can't tell what a credit buys, and Apple's "no guarantee of the same output" wording is that problem stated plainly.

Top-ups on top of a subscription, which is the Apple move. It keeps the base price low, but the moment someone needs one is usually the moment they're most invested in the task.

Bring your own API key. Popular with developers, close to unusable for everyone else.

Tiering by model quality instead of quantity. Cheaper model for everyday use, the expensive one on request.

For Murror, the thing I keep coming back to is when the limit shows up, not how big it is. Someone halfway through writing about a hard day should never see a usage warning. If there has to be a limit, I'd rather it show up at a calm moment, like a settings page or the start of a session, than in the middle of the thing the product exists for. I suspect that's true well beyond emotional products. A limit someone hits mid-flow feels like a paywall, even when it's the same limit.

So, an open question for makers here, and I'd love to see people compare notes with each other, not just with me:

How are you charging for AI features that cost you money every time they run? Flat with a cap, credits, top-ups, bring your own key, something else?

And has a pricing choice ever changed how people actually used your product, for better or worse?