AI·4 min read
Credits you can predict: pricing an AI run before it starts
Model calls are variable. Bills shouldn’t be. How a run stays inside a budget it agreed to up front.
An agent run can make dozens of model calls, each with a different prompt size and output length. Charging after the fact means surprise bills; refusing to start means a worse product. MoboStudio AI does neither.
Reserve, then spend
Before planning begins, a run reserves credits: the smaller of what’s available and a per-run maximum. From then on, no call can overspend. Before every model call, the run prices its prompt as if nothing were cached — the worst case — and caps the output, thinking included, to what the remaining reservation can pay for.
Stop gracefully
When the remaining budget can’t cover a meaningful call on the preferred model, the run pauses instead of calling. The work so far is kept; the user can send “continue” to resume. If the balance can’t even cover the planner’s first call, the run fails immediately — before any model is touched.
Settle honestly
- Credits are settled at actual cost, never above the reservation.
- Cached tokens are billed as cache reads where the provider reports them.
- Every charge is itemised on the usage page.
Predictability is a product feature. People build more when they aren’t afraid of the bill.