Free resource
LLM & voice AI cost worksheet
The same model behind the live cost estimator on the demos page, as a worksheet you can fill in by hand or copy into a spreadsheet.
Inputs
| Expected monthly active users | _______________ |
| Average voice minutes per user, per month | _______________ |
| Language (English adds no penalty; Arabic adds endpointing overhead) | _______________ |
| Voice quality tier (standard / premium synthesis) | _______________ |
| Average LLM tokens per turn (prompt + completion) | _______________ |
| Turns per session | _______________ |
Cost breakdown — fill in your provider's published per-unit price
| Component | Unit | Rate | Monthly cost |
|---|---|---|---|
| Speech-to-text | per minute | _______ | _______ |
| LLM inference | per 1K tokens | _______ | _______ |
| Text-to-speech synthesis | per 1K characters | _______ | _______ |
| Real-time transport / media relay | per minute | _______ | _______ |
| Hosting / compute | flat monthly | _______ | _______ |
| Total | _______ |
Per-user economics
| Total monthly cost ÷ monthly active users | = cost per user |
| Cost per user vs. planned price per user | = gross margin |
Arabic sessions typically add an endpointing penalty to turn latency, not directly to cost — but a longer average session (from re-prompts on cut-off speech) does raise cost per user. Model that separately if Arabic is a meaningful share of traffic.