AI voice agent cost: minutes, carriers, outcomes
Budget an AI voice agent as three costs: the voice engine, the phone carrier and the work needed to operate it. On Wixzel Voice, engine usage comes from one prepaid balance and telephony is billed separately by your SIP carrier. Compare cost per completed task as well as cost per minute.
Last updated
By Aqeel Shamsudheen, founder of Wixzel Voice. Published .
How do you calculate a monthly voice agent budget?
For an initial estimate, multiply connected calls by average connected minutes and the engine’s estimated per-minute rate. Add carrier charges, number rental and any other separately billed services. Then budget the engineering and support time needed to maintain prompts, integrations and failure handling.
For example, 1,000 connected calls averaging three minutes use 3,000 engine minutes. If only 400 calls connect, use those calls’ actual durations for the engine estimate; do not assume 1,000 attempts equal 1,000 conversations. Carrier billing rules and any minimum charges need their own calculation.
- Engine estimate = connected calls × average connected minutes × estimated engine rate.
- Carrier estimate = carrier-billed minutes × destination rate + number rental + applicable carrier fees.
- Operating budget = engine + carrier + other services + implementation and maintenance.
What does 1,000 three-minute AI calls cost on Wixzel Voice?
The table below uses 3,000 connected minutes at the published preset-engine estimates. These estimates already include orchestration: adding the platform fee again would count it twice. They exclude carrier fees, taxes and your team’s operating costs.
These are planning figures, not quotes or measured workload results. Token and character usage depends on how much the caller and agent speak, the prompt and the model. A real call’s itemized usage is the source for what was charged.
| Engine | Estimated per minute | 3,000 connected minutes |
|---|---|---|
| Classic | $0.0867 | $260.10 |
| Sarvam | $0.0623 | $186.90 |
| Gemini Live | $0.0466 | $139.80 |
| Deepgram Agent | $0.1851 | $555.30 |
USD, estimated at typical speech rates; orchestration included. Carrier fees, taxes and operating costs excluded. Actual usage varies.
Why measure cost per completed task?
A lower per-minute rate can still cost more per booking or resolved enquiry if calls take longer or need more retries. Define success before testing: for a booking, it might mean the requested appointment was recorded correctly and confirmed to the caller.
Divide all pilot spend, including failed attempts and retries, by independently verified successful tasks. In a hypothetical pilot costing $120 with 80 verified completions, the cost is $1.50 per completion. This is an arithmetic example, not a Wixzel Voice performance claim.
- Run the same scenarios and languages on each engine.
- Record connection rate, task completion, conversation duration, handoffs and total spend.
- Keep unsuccessful conversations in the cost numerator; exclude them from successful outcomes.
How do you check the estimate against the bill?
Run a small consented pilot, read each call’s cost_micros, and reconcile it with the usage events and your carrier’s bill. One million micro-dollars equals one US dollar. Use the actual average duration and actual spend from that pilot to revise the budget.
Wixzel Voice uses prepaid credit with a $5 minimum top-up and no subscription or free tier. Web and mobile audio sessions have no phone-carrier cost because they do not use a phone line. Self-hosting has a separate license, infrastructure and direct provider costs; it does not make inference or telephony free.
Frequently asked questions
- Does AI voice agent pricing include phone numbers?
- On Wixzel Voice, no. You keep your numbers and SIP trunk with your carrier and pay that carrier directly. The published engine estimates cover voice processing and orchestration.
- Is a 90-second call billed as two minutes?
- Wixzel Voice meters actual usage per second rather than rounding connected duration to whole minutes. Token and character components follow their usage units. Your carrier may use different billing increments.
- Can I use the cheapest engine for every language?
- Choose an engine that supports your required language, then test it on representative conversations. Price alone does not establish transcription accuracy, pronunciation or task completion.
Sources and product details
This resource is published by Wixzel Voice. Product details come from the references below; evaluation recommendations are a proposed method, not benchmark results.
Found an error? Send a correction.