If you're building voice AI agents for clients, you need to understand how billing really works. Because it's not as simple as "charge per minute."
Retell AI's billing model has hidden complexity:
The base model is per-minute pricing. But there are two exceptions most people miss:
Exception 1: Dynamic opening minimum.
If your AI speaks first and the call ends in under 10 seconds, you're still billed for 10 seconds. That 6-second hang-up? Billed as 10 seconds.
Exception 2: Long prompt surcharge.
When your total LLM context exceeds 3,500 tokens, your billed duration gets scaled up.
The formula: Billed duration = actual duration × (tokens ÷ 3,500)
A 2-minute call with 7,000 tokens? You're paying for 4 minutes.
The manual billing nightmare:
Retell shows usage at the workspace level, not per client. So if you're serving multiple clients from one account, you're stuck:
- Exporting logs.
- Calculating minutes and token scaling per client.
- Reconciling everything in spreadsheets.
- Creating invoices manually.
- Chasing payments.
This doesn't scale and is prone to errors.
Here's the thing: Professional voice AI agencies don't do any of this.
Want to know how? Check out my full breakdown in the article.