Controlling Spending and Usage

Usage-based billing worries people who have never used it, almost always because of one fear: waking up to a bill they did not expect. That is worth addressing directly, because the structure makes it very hard to happen.

The balance is prepaid

This is the most important thing on the page. You are not on an account with a monthly invoice that arrives after the fact: you add credit first, and the tools draw it down. When the balance reaches zero the features stop working until you top up again.

The most you can spend is the amount you have already put in. Start with a small balance while you learn what your usage looks like.

Set limits at the provider

Both dashboards let you go further than the balance itself:

• Spending limits: cap how much a key may consume over a period.
• Per-key separation: issue one key per machine or per user, so you can see who is spending what and revoke one without disturbing the others.
• Usage history: a breakdown by model and by day, which is the honest answer to “where is the money going”.

Choose models deliberately

The gap between a cheap model and a frontier one is not a rounding error, it is orders of magnitude. Which means the model dropdowns in Integrations are the biggest cost lever you have, bigger than any change in how much you use the tools.

See Quality, speed and cost for how to pick.

Habits that cost less

• Start a new conversation when you change subject. Every message in a thread carries the earlier turns as context, so long threads get more expensive per message. A fresh conversation resets that.
• Explore on a cheap image model, finish on an expensive one. Most of your generations are drafts you will discard. Pay the top rate only for the keeper.
• Write the prompt properly the first time. Ten vague attempts cost ten generations. One specific prompt often costs one.
• Ask the Assistant precise questions. A focused question gets a short answer, and you pay for the answer too.

In a studio with several seats

Give each machine its own key rather than sharing one. You get per-seat usage figures, you can set different limits, and if someone leaves you revoke a single key instead of rotating one that everybody is using.