Vapi Pricing: Vapi AI Pricing, Vapi Cost per Minute and Success Packages Compared

Vapi costs $0.05 per minute for its own platform, plus whatever the transcriber, the language model and the voice cost for that minute. There is no monthly fee to start: usage only begins at $0 with $5 of free credit, and you prepay minutes as you go. Every package, including usage only, runs on the same platform with, in Vapi's words, "no cap on agents and minutes."

What you buy on top is a success package. Vapi describes the split plainly: usage is "your Vapi hosting fee ($0.05/min) plus the model costs for each minute your agents run," while a package is "the layer around those minutes: support, SLAs, concurrency, and compliance." You can change one without touching the other, so a team can run a million minutes on Core or a thousand on Pro.

If you have read a Vapi pricing guide that talks about a Build plan and a Scale plan, it is describing an older price list. The current pricing page lists no plan by either name. It lists usage only, Core, Pro and Premier, and moves concurrency and organizations into add-ons you can attach to any of them.

Four things are in every minute: speech to text, the language model, text to speech and transport. Vapi's own FAQ says each is "billed at the provider's listed price with no markup," and that its $0.05 hosting fee "is added on top." So the per-minute price of a Vapi agent is a sum you control by choosing providers, not a single rate you negotiate.

Vapi's pricing calculator shows the ranges for its default stack. These are the figures it displayed on September 17, 2026:

The model is the swing factor. At the cheap end of OpenAI's range the whole minute costs about 8 cents, and Vapi's hosting fee is more than half of it. At the expensive end the minute costs about 13 cents and the model alone is a third of the bill. Longer system prompts and more tool calls push the model cost up, because Vapi's cost docs estimate it from input and output tokens per minute and the number of model requests per minute.

Two smaller items catch teams out. Vapi transcribes both sides of the call, so its cost docs note that the transcription estimate "includes two audio channels." And if you bring your own Twilio numbers instead of using Vapi's, add $0.008 a minute inbound or $0.014 outbound, which is a 10% to 17% increase on a cheap stack.

Here is what the default stack costs per month on usage only, using the calculator's low and high ends. The 1,000 minute row matches Vapi's own calculator estimate of $82 to $129.

Read the last column carefully. Pro is priced at "10% of Vapi hosting fee" with a "$999/mo minimum," and Vapi confirms that "model costs are not included" in the percentage. Ten percent of hosting equals $999 only when hosting reaches $9,990 a month, which is 199,800 minutes. Below that, and that covers most US teams, Pro is a flat $999 a month. At 10,000 minutes it nearly doubles the bill; at 100,000 minutes it adds 8% to 12%.

How many minutes will a US business actually use?

A front desk that takes 60 calls a day at an average of three minutes uses about 5,400 minutes a month over a 30 day month. A small contact center line with 300 calls a day at four minutes uses about 36,000. Outbound campaigns are harder to predict because unanswered dials and voicemail drops also consume time, so size them from a pilot week rather than a spreadsheet.

Buy a package for support terms and data retention, not for concurrency. Concurrency is an add-on at $10 per line per month on any package, including usage only, and Vapi says add-ons exist "so you never have to move up a tier just to get one thing you're short on."

The arithmetic is lopsided. Core with 20 extra lines gives you the same 30 concurrent calls as Pro for $229 instead of $999. What the extra $770 buys is a one business day response commitment, a dedicated Slack channel, 180 days of raw data retention instead of 30, role-based access control and the 99% uptime SLA. For a phone line that books revenue, that SLA can be worth it. For an internal pilot, it usually is not.

What does Vapi Core include?

Core costs $29 a month, prepaid and auto-renewing, and includes 2 organizations, 5 Vapi phone numbers, 10 concurrent calls, 30 days of raw data retention, zero data retention (ZDR) and email support with a two business day response SLA. Minutes are billed separately at the same $0.05 plus models. It is the package most small US teams should start on, because ZDR is only "included with the addition of any success package."

What does Vapi Premier include?

Premier is sold directly by sales with no public price. It adds a named support team with an Agent Strategist and a Forward Deployed Engineer Advisor, a 1 hour response for P0 incidents, a 99.9% uptime SLA, custom RBAC, ZDR and SSO, and dedicated SIP. Vapi's FAQ says the sales team will "walk through your expected volume, compliance requirements, and how much hands-on support you want" before recommending a package.

Voice agent platforms price the same minute in three different ways: a platform fee plus pass-through models (Vapi, Retell), one bundled rate that already includes the model and voice (Bland, ElevenLabs), or an annual enterprise contract (Synthflow). The table puts all five on one footing at 10,000 US minutes a month. Every rate was read from the vendor's own pricing page on September 17, 2026.

Three conclusions hold up. First, on a cheap model Vapi is the lowest-cost platform with a pass-through design, because its hosting fee is $0.005 below Retell's infrastructure fee and Vapi's own numbers carry no telephony charge. Second, Retell includes 20 concurrent calls at no charge, which Vapi would bill as usage plus 16 extra lines, $160 a month. Third, Bland's Start plan caps you at 100 calls a day. At three-minute calls that is about 9,000 minutes over a 30 day month, so a 10,000 minute line has to move to Build, and Build only beats Start on price above 14,950 minutes a month.

Is Vapi cheaper than Retell?

Usually yes on the per-minute bill, and not always on the total. With the same model and voice, Vapi's platform fee is $0.050 against Retell's $0.055 plus $0.015 for Retell's platform voice, and Vapi numbers carry no transport fee. But Retell includes 20 concurrent calls and $10 of free credit, and Vapi includes 4 concurrent calls and $5. A small team with bursty traffic can pay more on Vapi once it buys lines. For a line-by-line version of this comparison, with the other alternatives, see our guide to Vapi alternatives priced per minute.

HIPAA-eligible data handling on Vapi costs $2,000 a month per organization, on any package, and "comes with a standard DPA and BAA." That is the number US healthcare, dental and behavioral health teams need to budget, and it is fixed: at 5,000 minutes a month it is more than the minutes themselves. A clinic that runs separate dev and production organizations and wants both covered should confirm with Vapi whether both need the add-on, because the price is quoted per organization.

For enterprise buyers the path is Premier, which adds SSO, custom RBAC, dedicated SIP and the 99.9% uptime SLA, with the price set by sales. Before you sign, get three numbers in writing: the effective hosting rate per minute at your committed volume, the concurrency ceiling included in the package, and the data retention period for recordings and transcripts. Retention matters for compliance reviews: usage only keeps raw call data for 14 days, Core for 30 and Pro for 180.

If you are pricing a voice agent for a medical practice, our page on AI agents for healthcare covers the intake, scheduling and billing jobs where a ready-made agent can take load off the phone line.

Vapi is worth it when you have engineers who want to own the stack. You pick every model and voice, you pay providers at cost, and you can swap components when prices drop. That is real control, and at scale the 5 cent hosting fee is a fair price for not building call orchestration yourself.

It is the wrong purchase when nobody on the team wants to build. Vapi is a developer platform: the price list assumes you will design prompts, wire tools, test calls and monitor them. The honest total for a production line includes that engineering time, which dwarfs a few hundred dollars of minutes.

Many calls also do not need a voice agent at all. If a large share of your inbound calls are "where is my order," refund requests or the same ten support questions, deflecting them to chat and email is often cheaper than answering them by the minute. That is what the ready-made agents in our catalog do: a support agent that resolves common tickets from your own help docs, an order concierge, a refund resolver. Each listing shows its flat monthly price, its integrations and the exact permissions it asks for before you deploy. Start with our AI agents for customer service, or compare how Sierra AI prices each resolution and what Fin charges per resolution if you are weighing an outcome-priced support agent instead.