Voice Agent Cost Calculator
Price the same call minute across ElevenLabs Agents, Vapi, Retell, Bland, Synthflow, a human answering service, and a self-hosted stack.
Voice agent platforms quote per-minute rates that are not comparable. Some bundle speech, language model, voice, and telephony into one number. Others quote a thin platform fee and bill every component on top. The headline rate tells you almost nothing until you model your own call volume.
Enter your monthly calls and average call length below. The calculator prices the same workload across every option and shows the volume at which self-hosting starts to make sense.
Compare Voice Agent Costs for Your Call Volume
Turn this off to compare headline platform fees only. Headline fees make bring-your-own-keys platforms look cheaper than they bill.
2,400 conversation minutes a month. Lowest monthly cost here: Retell AI at $168.
| Option | Rate used | Monthly | Annual | How it bills |
|---|---|---|---|---|
| Retell AI Runs roughly $0.07 to $0.31 per minute depending on the components selected. The low end is used here. | $0.07 | $168 | $2,016 | Pay-as-you-go per minute, no separate platform fee |
| ElevenLabs Agents Around $0.08/min on annual business plans, about $0.10 on Creator and Pro. The rate covers the voice layer only. | $0.13 | $312 | $3,744 | Voice per minute; LLM and telephony billed separately |
| Bland AI From about $0.14/min on Start, falling on committed monthly tiers. No separate STT, LLM or TTS keys. | $0.14 | $336 | $4,032 | All-inclusive per connected minute |
| Vapi A $0.05/min platform fee. Bring your own keys, so the fully-loaded figure depends entirely on the components you pick. | $0.18 | $432 | $5,184 | Thin platform fee; STT, LLM, TTS and telephony at cost on top |
| Self-hosted stack A GPU host plus telephony trunking, with engineering time excluded. The marginal minute is close to free; the fixed cost is not. Engineering and on-call are the real expense and are NOT in this figure. | $0.01 + $900/mo fixed | $924 | $11,088 | Fixed infrastructure, near-zero marginal minute |
| Human answering service Live answering services commonly bill per minute in the low single dollars, or $2,000-$4,000+/mo fully loaded for an in-house receptionist. | $1.35 | $3,240 | $38,880 | Per minute, human staffed |
| Synthflow Synthflow no longer publishes a per-minute rate. Its pricing page states enterprise contracts start at $30,000 a year, scoped on call volume, concurrency, telephony and integrations. Verified 2026-08-22. We will not estimate a per-minute figure the vendor does not publish - get it in writing. | No public rate | Quote only | Quote only | Enterprise contract, quoted per deployment |
Self-hosting overtakes the cheapest managed option at roughly 15,000 minutes a month. That figure covers infrastructure only. Engineering time, on-call, and model updates are real costs and are not included, so treat it as a floor, not a business case.
Rates last checked 2026-08-22. Voice-agent pricing changes without notice and several vendors have cut rates mid-year. This is a directional estimate, not a quote: confirm every figure on the vendor pricing page before you budget. Rates shown are entry public pricing; committed and enterprise tiers are lower.
How We Calculate Your Numbers
Every rate in this calculator is copied from a vendor's own public pricing page. Nothing is averaged, inferred, or estimated by us, and each row links to our full breakdown for that platform.
- Monthly minutes: your calls per month multiplied by average call length.
- Monthly cost: monthly minutes multiplied by the rate for that platform, either the headline fee or the fully-loaded rate.
- Self-hosted cost: a fixed monthly infrastructure figure plus a near-zero marginal minute. Engineering time and on-call are excluded and are the real expense.
- Crossover volume: the fixed self-hosting cost divided by the gap between the cheapest managed per-minute rate and the self-hosted marginal minute.
This is a directional estimate, not a quote. Committed and enterprise tiers are lower than the public rates shown, and every vendor here has changed pricing before.
Bundled vs Pass-Through Pricing
The single biggest cost mistake is comparing a bundled rate against a platform fee. A bundled platform quotes one number that covers speech recognition, the language model, the voice, and the phone line. A pass-through platform quotes only its own margin.
A five cent platform fee can land well above a ten cent all-in rate once the language model and telephony arrive on the bill. That is why the fully-loaded toggle is on by default. Turn it off only when you are deliberately comparing headline fees.
Pass-through pricing is not worse. It buys you the freedom to swap any component and to negotiate each one as volume grows. It just makes forecasting harder in year one.
When Self-Hosting Starts to Win
Self-hosting wins on marginal cost and loses on fixed cost. A GPU host and telephony trunking cost the same whether you run one call or ten thousand, so the per-minute cost falls as volume rises.
The crossover figure above covers infrastructure only. The honest version includes an engineer who owns the stack, keeps models current, and answers the phone at 2am when a deploy breaks call routing. For most businesses that person costs more than the per-minute savings.
Self-hosting makes sense when call volume is high and steady, when data residency rules out a managed vendor, or when you already run GPU infrastructure for something else. See our self-hosted voice agent stack guide for the full architecture.
Frequently Asked Questions
- ElevenLabs Agents bills around $0.08 per minute on annual business plans and about $0.10 on Creator and Pro tiers. That rate covers the voice layer only. The language model driving the conversation and your telephony are billed separately on top, so your real per-minute cost is higher. Confirm current rates on the ElevenLabs agents pricing page before budgeting.
- Because platforms bill on two different models. Bundled platforms quote one all-in rate. Pass-through platforms quote a thin fee and bill speech, language model, voice, and phone at cost on top. Toggle fully-loaded rates on to compare what you will actually pay, and off to compare headline fees.
- It depends on your call volume and length, which is why this is a calculator rather than a table. At low volume, a bundled all-in rate is usually cheapest and easiest to forecast. At high volume, a pass-through platform where you negotiate each component, or a self-hosted stack, usually wins.
- Self-hosting overtakes managed platforms once your fixed infrastructure cost divides down below the per-minute rate you would otherwise pay. The calculator shows that crossover for your inputs. Treat it as a floor: it covers infrastructure only and excludes engineering time and on-call, which are the real expense of running your own stack.
- Every rate is copied from the vendor's own public pricing page and carries the date we last checked it, shown under the table. Voice-agent pricing has moved repeatedly and several vendors have cut rates mid-year. Always confirm on the vendor page before committing budget.
- No. This calculator covers running cost per minute only. Build effort varies enormously — a no-code agent can launch in a day, while a custom integration with your CRM, calendar, and warm-transfer rules takes real engineering. Factor that in separately.
Not sure which platform fits your call volume?
Layer3Labs builds custom voice agent systems on whichever platform actually fits your economics — or on your own infrastructure when the volume justifies it. We have no vendor quota.
Book a Free AI Workflow Audit