AI Receptionist: How It Works and What It Costs Per Minute

What does an AI receptionist actually do on a call?

It answers, listens, understands and acts. Four things happen in under a second: the telephony layer carries the call, a transcription engine turns speech into text, a language model decides what to say and what to trigger, and a synthetic voice speaks the answer back. Depending on the flow you write up front, the agent asks a qualifying question, transfers to an extension, books a slot or takes a message.

That chain is exactly why the pricing looks the way it does. You are not buying a piece of software, you are renting a stack of providers metered by the minute. Vapi builds its whole product around that idea and lets you choose each part yourself, from the model to the voice to the transcription; it scores 3.7 out of 5 on Stacknota. Retell AI pre-assembles more of the stack, at 3.6 out of 5.

What makes up the price of one minute?

Every block has its own meter, and vendors do not publish the same level of detail.

The Retell AI range says most of what you need to know: from one stack to another, the same minute can cost roughly four times as much. With Vapi, the $0.05 per minute is only one slice of the total, since models are rebilled with no markup, which we count as a strength and a friction at the same time. You assemble the total yourself.

What do Retell AI, Vapi and Synthflow publish?

With Vapi, the Pro plan is 10% of your Vapi hosting fees with a floor of $999 per month, and it brings 30 concurrent calls, 180 days of retention, role-based access control and a 99% uptime SLA. The gap between Core at $29 and that floor is the sharpest edge in the lineup.

Synthflow plays a different game: no self-serve plan anymore, a $30,000 per year entry ticket, and a price set by call volume, concurrency, telephony, integrations and security requirements. In exchange, the vendor runs the launch with you and offers native or SIP telephony. Its 1.5 out of 5 on price reflects that entry ticket, not the quality of the platform. For a feature-by-feature look at the three, see our comparison of AI voice agent platforms.

What pushes the bill above the per-minute rate?

Three line items tend to slip past the budget. Concurrency first: Retell AI includes 20 concurrent calls, then charges $8 per month for each additional one, while Vapi gives you 4 with no plan, 10 on Core at $29 per month and 30 on Pro. Phone numbers next: $2 per month at Retell AI, 1 number included with no plan at Vapi and 5 on Core. Compliance last: at Vapi, the HIPAA add-on is billed at $2,000 per month.

In our reading, the number that decides your budget is not the headline per-minute rate, it is the concurrency you need at peak. A clinic that gets ten calls at once on Monday morning sizes its plan on that spike, not on its average volume.

Where does your call data live?

This is the widest split between the three, and the one that matters most if your contracts touch Europe.

Retell AI processes and stores data in the United States on AWS, and its documentation states that it does not yet operate services within the European Union. The DPA, with EU standard contractual clauses, can be signed online at no cost, and retention is configurable per agent from 1 day to 2 years. Certifications listed: SOC 2 Type 1 and Type 2, plus HIPAA. GDPR score: 2.5 out of 5.

Vapi runs mainly on AWS with deployments in the US and the EU, each organization belonging to a single region, and documents its EU region explicitly. The Trust Center lists SOC 2 Type II and SOC 3 for the July 2025 to July 2026 period, plus HIPAA, GDPR, CCPA and PCI DSS v4.0.1. GDPR score: 4.5 out of 5.

Synthflow shows the longest certification list of the three (SOC 2 Type 2, ISO/IEC 27001, ISO/IEC 42001, HIPAA, GDPR and PCI DSS v4.0.1) with a DPA on request, but does not publish where the data sits. Plenty of labels, one missing fact, and that is the question to put in writing before signing.

Which one fits your situation?

If you want to test fast with no commitment and US hosting raises no contractual problem, Retell AI is the shortest path: no subscription, $10 in credits at sign-up, 20 concurrent calls included and a component-by-component price list. Details are on our Retell AI page.

If your contract or your industry requires processing inside the European Union, Vapi is the only one of the three that documents an EU region. You can still start with no plan and $5 in credits, but budget engineering time: we score it 2.5 out of 5 on ease of use, because every block is assembled through the API. The Core plan at $29 per month adds zero data retention mode and 30 days of retention. Full breakdown on our Vapi page.

If you run a contact center with an annual budget already approved and you want a guided rollout, Synthflow matches the need, provided you accept the $30,000 per year floor and the absence of any published per-minute rate.

How do you test before committing a budget?

Start with one scenario you can measure: appointment booking, or qualifying inbound calls. Use the trial credits ($10 at Retell AI, $5 at Vapi) to measure your real cost per minute on your own stack rather than reasoning from a published range. Then test the handoff to a human, which is where these agents earn or lose their keep day to day.

On the data side, get the DPA before going live and set retention to the shortest window your business actually needs. Retell AI allows 1 day at the low end; Vapi keeps raw data 14 days with no plan. Finally, if the agent has to write into your business tools, scope that integration on day one: that is usually where setup time goes, far more than the call minute itself. Our guide on building an AI agent covers the orchestration side in detail.