AI inference cost calculator
The single most common founder question we get in DMs: 'what will this actually cost in production?' This calculator answers it across eight widely used models, with current public pricing.
Daily
$6.00
Monthly
$180
Yearly
$2,190
The unit-economics question
If you’re building an AI product, your COGS is inference cost. The biggest lever is routing: send the bulk of calls to cheap models and reserve reasoning models for the work that needs them. Wend runs its production stack on Azure-hosted models plus open models behind a router, and the Mac app routes short extraction jobs to Apple’s on-device model where it is available, so per-user inference cost stays well below the gross margin we’re willing to give up.
For a feature-by-feature breakdown, the most useful discipline is: every call gets a model class (“cheap”, “reasoning”, “vision”), not a model name, so you can swap models without rewriting code. We wrote about that pattern in the audit-log post linked above.