AI Voice Agent Pricing: $0.05–$0.31/min Across 4 Models (2026)

TL;DR: Published AI voice pricing starts at Vapi’s $0.05 platform fee, Retell’s $0.07–$0.31 pay-as-you-go range, Bland’s $0.11–$0.14 connected-minute rates plus plan fees, or Synthflow’s $30,000 annual enterprise minimum. Compare one fully loaded call—not headline rates—and verify telephony, models, transfers, QA, implementation, and support.

Direct answer: There is no single “AI voice agent price.” Developer platforms separate orchestration, models, speech, and telephony; bundled platforms meter connected time; enterprise providers scope annual contracts. At 5,000 connected minutes, the published inputs below span $250 for Vapi’s platform layer alone, $350–$1,550 across Retell’s stated pay-as-you-go range, and $700–$1,049 on Bland’s self-serve tiers before optional extras. Terms were checked against official vendor pages on July 31, 2026.

The table compares public billing inputs, not call quality or total cost of ownership. It does not rank the vendors. A lower meter can still produce a higher business cost if implementation, failed-call review, or human escalation takes more labor.

What Does Each AI Voice Pricing Model Actually Include?

Vapi: a platform layer with provider costs passed through

Vapi’s official pricing page lists a $0.05-per-minute Build platform fee. The same page says model costs are passed through to the customer and exposes separate choices for transport, speech-to-text, large language model, and text-to-speech. Ten concurrent calls are included; additional concurrency is listed at $10 per line per month. HIPAA and Zero Data Retention appear as separate monthly add-ons on the public page.

That makes $0.05 a valid platform price, but not a valid all-in call price. A buyer needs to select the model, voice, transcription, and telephony stack before calculating the actual minute. At 5,000 minutes, $250 is only the Vapi platform layer:

5,000 minutes × $0.05 platform fee = $250
Fully loaded cost = $250 + model + TTS + STT + transport/telephony + add-ons

Vapi is easiest to compare when the buyer already knows the provider mix and can estimate each line separately.

Retell: a visible component calculator and a wide total range

Retell’s official pricing page publishes $0.07–$0.31 per minute for pay-as-you-go Voice AI. It also breaks out the current components: $0.055/minute for voice infrastructure, voice pricing by provider, model pricing by model and speed tier, and $0.015/minute for Retell telephony. Custom telephony can remove that Retell telephony line.

The page also lists optional per-minute or per-action add-ons, including knowledge base, batch calls, branded calls, denoising, guardrails, personal-information removal, and AI quality assurance. Twenty concurrent calls are included; added concurrency and Retell phone numbers have monthly prices.

The useful comparison is a configured estimate, not the top or bottom of the range. At 5,000 minutes:

Low published range: 5,000 × $0.07 = $350
High published range: 5,000 × $0.31 = $1,550

A buyer should save the calculator configuration with the quote so a later model or voice change does not silently invalidate the budget.

Bland: plan fee plus connected-minute and transfer rates

Bland’s billing documentation says its current self-serve structure took effect December 5, 2025:

Bland bills connected time to the second. Its documentation says transfer time is billed additionally when a call transfers through a Bland-provided number; bring-your-own-Twilio transfers do not carry that Bland transfer fee. Enterprise pricing is custom.

At 5,000 connected minutes before transfer charges:

Start: 5,000 × $0.14 = $700
Build: $299 + (5,000 × $0.12) = $899
Scale: $499 + (5,000 × $0.11) = $1,049

The lower Scale minute rate does not beat Start at this volume because the monthly fee is larger. Solve the full equation before upgrading solely for the meter.

Synthflow: an enterprise annual commitment

Synthflow’s current pricing page says enterprise contracts start at $30,000 annually. It says final pricing is scoped around call volume, concurrency, telephony setup, integrations, security needs, and launch support.

That is not directly comparable to a self-serve platform fee. The annualized floor is $2,500 per month, but the page does not promise a public minute allowance at that price. A fair comparison requires a written scope showing included usage, implementation, support, overages, and contract term.

How Do You Calculate the Fully Loaded Cost per Minute?

Use one formula for every quote:

Monthly platform or contract fees
+ connected-minute charges
+ telephony and transfer charges
+ model, speech, and voice charges
+ numbers, concurrency, storage, and compliance add-ons
+ implementation and ongoing QA labor
= fully loaded monthly cost

Fully loaded monthly cost ÷ connected minutes
= fully loaded cost per connected minute

Do not divide by attempted calls. Failed calls, voicemail, transfers, and silence can follow different billing rules. Use the vendor’s billing definition and your own call logs.

Prestyj’s current planning benchmark puts managed, high-volume implementations at $0.06–$0.18 per minute at 50,000+ minutes per month. That is a first-party benchmark under stated scale assumptions, not a promise that every account or vendor reaches the range. Low-volume pilots should expect a higher effective cost because setup and review labor are spread over fewer calls.

Which Hidden Costs Should Buyers Put in the Quote?

A useful quote answers these questions in writing:

1. What starts and stops the clock? Ask about ringing, silence, hold time, voicemail, transfer time, and post-transfer billing.
2. Which model and voice produced the demo? A low-cost configuration may not be the one used in the sales call.
3. Is telephony included? Separate carrier, SIP, phone-number, or transport charges can change the total.
4. What happens above included concurrency? A campaign can fail operationally even when the per-minute budget is correct.
5. Which add-ons are required in production? Knowledge bases, guardrails, redaction, call recording, retention, branded calling, and QA may be optional line items.
6. Who owns implementation? Prompt design, call routing, CRM actions, calendar rules, testing, and escalation need labor even when software setup is free.
7. Who reviews failures after launch? Budget time for transcripts, call recordings, false transfers, missed intents, and knowledge updates.
8. What is the contract minimum? Annual commitments and committed volume can dominate the meter.

Prestyj’s first-party pricing synthesis estimates hidden charges and omitted operating lines at 18–35% above the advertised monthly price across the contracts it reviewed. Treat that as a planning allowance until a vendor-specific quote replaces it.

What Should an AI Voice Pilot Cost?

The software bill is only one part of a pilot. A proper pilot also needs a bounded use case, success criteria, test calls, human escalation, consent and recording review, and a rollback path.

Prestyj’s current SMB planning range is $0–$1,500 in setup fees for a scoped pilot with persona configuration, knowledge ingestion, a CRM webhook, and call routing. The range does not include every vendor’s usage charges, your internal labor, or enterprise security work.

Use a 30-day pilot scorecard:

A cheap call that books the wrong appointment or mishandles an emergency is expensive. A higher-priced call that resolves the workflow safely can be cheaper at the business level.

Which Pricing Model Fits a Service Business?

Default to the model that matches your team’s operating capacity:

• Choose a developer platform when engineers want provider-level control and will own monitoring, integrations, and incident response.
• Choose a bundled self-serve platform when an operator can build and maintain workflows but wants fewer provider bills.
• Choose an enterprise contract when security, support, concurrency, procurement, and launch assistance matter more than self-serve flexibility.
• Choose a done-for-you service when the business wants a managed outcome and does not have a person available to own prompts, routing, QA, and optimization.

The commercial decision should use cost per resolved call or cost per booked qualified appointment—not cost per minute alone. Compare the same workflow, call mix, quality threshold, escalation rules, and labor assumptions.

What Changed in This July 2026 Refresh?

This page previously showed stale or unsupported plan tables for seven vendors, including old Bland rates, a discontinued public Synthflow self-serve structure, unverifiable Air.ai packages, and obsolete Prestyj voice plans. Those tables were removed.

The replacement uses only pricing visible on official Vapi, Retell, Bland, and Synthflow pages checked July 31, 2026. Public pricing can change without notice; verify the linked vendor page and save the quote date before buying.

What Is the Bottom Line on AI Voice Agent Pricing?

Start with the official meter, then add every component required to run one production call. Vapi’s $0.05 is a platform layer. Retell publishes a configured $0.07–$0.31 range. Bland combines a plan fee with $0.11–$0.14 connected minutes. Synthflow starts with a $30,000 annual enterprise commitment.

Normalize all four into the same workflow and calculate the fully loaded cost per resolved call. Then run a bounded pilot with human escalation and a written QA threshold.

For deeper buying research, read AI voice agent pricing, hidden costs of AI voice agents, AI voice agent cost per minute at scale, and voice agent testing pricing.

Prestyj delivers done-for-you AI agents for marketing and sales for service businesses and real estate teams. Book a demo to scope one call workflow using your actual volume, escalation rules, and success metric.