Vapi Pricing (2026): $0.05/Min Is Only the Platform Fee

Vapi starts at $0.05 per call minute, but that is only the hosting layer. For a production voice agent, Vapi beats Retell AI's live $0.115 example only when your combined STT, LLM, and TTS bill stays below $0.065 per minute, before telephony and fixed add-ons.

Vapi pricing at a glance

Vapi is a developer platform for assembling voice agents from separate speech, language, voice, and phone providers. Its Build plan is cheap to enter and flexible to configure, but the price on the front door is not the finished-call price.

Prices and limits were verified against Vapi's live pricing page and current documentation, August 2026.

Build passes speech-to-text, language-model, and text-to-speech costs through at cost. Bring your own provider key and Vapi shows $0 for that provider line, but the provider bills you directly. Moving a charge to another invoice does not remove it.

Scale replaces public unit pricing with a negotiated annual contract. Vapi lists committed volume, enterprise uptime, SSO, RBAC, data residency, priority provider access, a support SLA, and a dedicated account team, but it does not publish the discount curve. You need a quote before you can compare Scale with self-serve economics.

The $0.05 fee is a floor, not the bill

The five-cent floor is Vapi's orchestration charge. It pays for the layer that keeps a call moving between the caller, transcription, the language model, voice generation, and the phone network. A production minute still has five budget lines:

• Vapi hosting: $0.05 per call minute.
• STT: speech-to-text turns the caller's audio into words and is passed through at cost.
• LLM: the language model decides what the agent says and is passed through at cost.
• TTS: text-to-speech turns the reply back into audio and is passed through at cost.
• Transport: the phone carrier or SIP provider moves the call and bills separately.

Vapi's own calculator exposes those layers separately. That is good architecture because a technical team can swap providers, bring negotiated API rates, or use its own keys. It is less comfortable for a finance owner who wants one predictable number per connected minute.

The honest formula is straightforward:

Vapi all-in cost per minute = $0.05 hosting + STT + LLM + TTS + transport + fixed add-ons divided by usage.

Vapi's provider-key documentation makes the invoice split explicit: once a key is validated, Vapi stops charging for that provider and the provider charges the account directly. This is useful for control and volume agreements. It also means your Vapi dashboard alone is not a complete cost ledger.

What a Vapi call costs at working usage

Vapi's hosting math is simple enough to budget before choosing a single model or voice. One 4-minute call costs $0.20 in Vapi hosting. At 1,000 call minutes the hosting bill is $50; at 5,000 it is $250; at 10,000 it is $500; at 50,000 it is $2,500. Provider and transport charges sit on top of each figure.

For a receptionist workflow, cost per outcome is more useful than cost per minute. Assume calls average 4 minutes and 20% of connected callers book. Each booking consumes 20 call-minutes across the successful and unsuccessful conversations, so Vapi hosting contributes $1 per booked appointment before providers and telephony.

That gives you a reusable decision rule. If STT, LLM, and TTS together cost $0.065 per minute, Vapi reaches Retell AI's $0.115 live example before telephony. If those providers cost $0.09 per minute, Vapi reaches Bland AI Start's $0.14 bundled AI rate before telephony. Provider selection is therefore a budget decision, not just a latency or voice-quality decision.

The hidden costs that change the budget

Concurrency, compliance, and billing controls can matter more than the $0.05 usage fee. They are fixed or peak-driven costs, so dividing them by an average monthly minute count can hide the moment your economics break.

Concurrency prices the peak, not the average

Vapi includes 10 simultaneous call slots. Each reserved line above that costs $10 per month. A business that needs 20 simultaneous calls therefore pays $100 per month for 10 extra lines.

At 1,000 monthly minutes, that capacity adds $0.10 to every minute. At 10,000 minutes, it adds $0.01. The same peak requirement is punishing for a low-volume service desk and almost invisible for a high-volume outbound operation.

Vapi's concurrency guide says excess calls wait when all slots are occupied and recommends discussing custom plans when usage regularly exceeds 50,000 minutes per month. Averages do not protect a receptionist from a Monday-morning spike. Model the busiest interval.

Compliance creates a volume hurdle

HIPAA costs $2,000 per month and Zero Data Retention costs $1,000 per month on both Build and Scale. Buy both and the $3,000 fixed charge contributes $3.00 per minute at 1,000 minutes, $0.30 at 10,000, and $0.06 at 50,000.

Add Vapi hosting and the pre-provider floor becomes $3.05, $0.35, and $0.11 per minute at those same volumes. A small clinic does not get enterprise compliance economics merely because the base API fee is usage based.

Build retains call history for 14 days and chat history for 30 days. If your policy requires shorter retention, Zero Data Retention is not a small checkbox. It is a separate budget line whose effective rate depends on volume.

Credits and traffic need guardrails

Vapi's billing guide sets a $10 minimum for manual credit purchases and auto-reloads. Saving an auto-reload rule charges the card immediately when the current balance is already at or below the chosen threshold.

Neither that guide nor the current pricing page publishes an expiration period for purchased credits. If you plan to prepay a material balance, get the expiry treatment in writing instead of assuming the credits remain available indefinitely.

The more serious risk sits in Vapi's terms: unless you limit traffic, incoming calls are not stopped merely because usage exceeds what you expected, and you remain responsible for the minutes used. Put rate limits, alerts, and a deliberate reload threshold in place before publishing a number.

Scale is an annual-lock decision

Build is pay as you go. Scale is an annual contract with committed volume and a custom rate. That can be the right move above 50,000 monthly minutes or when enterprise controls are mandatory, but a discount is not savings if the commitment exceeds your connected traffic.

Ask sales for the fixed platform fee, committed minutes, overage rate, concurrency, provider treatment, support terms, and renewal mechanics in one quote. Compare the total annual obligation with twelve months of Build at your conservative volume forecast.

Is Vapi free?

Vapi is free for a prototype, not for a live phone operation. The Build page advertises 60+ included call minutes, then charges by usage and passes model costs through. It does not publish a recurring monthly free allowance in the pricing-page text.

Vapi also offers a managed US phone number for testing. Its free telephony guide says one number can be requested without adding a payment method, accounts can hold up to five free Vapi numbers, and international numbers are not free.

Who never needs to pay? A builder who stays inside the initial allowance, uses the prototype briefly, and incurs no separate provider bill may finish an evaluation without paying Vapi. Anyone answering customer calls, making outbound calls, using paid providers, or operating beyond that trial window should budget from the first production minute.

There is no public student discount on Vapi's current pricing page or in its terms. The free allowance is the honest entry path for a student project.

Vapi vs Retell AI vs Bland AI

The right comparison normalizes what each published minute includes. For a broader product view, see the best AI voice agent platforms and the separate buyer guide to AI receptionists.

Retell AI: the cleaner benchmark for included concurrency

Retell AI publishes a $0.07 to $0.31 per-minute pay-as-you-go range and gives new accounts $10 in free credits. Its live calculator currently shows $0.115 per minute: $0.045 for the LLM, $0.055 for voice infrastructure, and $0.015 for TTS, with $0 for telephony and add-ons in that example.

Retell includes 20 concurrent calls, twice Vapi Build's allowance. Extra capacity costs $8 per line per month after that, and a Retell phone number costs $2 per month. Billing is tracked to the nearest second; silence and hold time are billable, failed calls are not, and voicemail is billed only while the agent remains active.

The decision flips at $0.065 per minute of Vapi provider cost. Below it, Vapi's $0.05 hosting layer beats Retell's $0.115 example before telephony and fixed add-ons. Above it, Retell's example is cheaper. Retell also wins sooner when you need 20 simultaneous calls because Vapi charges $100 per month to add the missing 10 lines.

Retell's limitation is its wide published range. $0.115 is a configuration example, not a universal invoice. Premium models, voices, telephony, and add-ons can move the total toward the top of its $0.07 to $0.31 range.

Bland AI: the predictable bundled rate

Bland AI bundles the LLM, STT, and TTS into its per-minute AI rate. Start costs $0.14 per minute with no platform fee, 10 concurrent calls, 100 calls per day, and 10 knowledge bases. Telephony remains separate at carrier or pass-through cost.

Build costs $0.12 per minute plus $299 per month and raises capacity to 50 concurrent calls and 2,000 calls per day. Scale costs $0.11 per minute plus $499 per month, with 100 concurrent calls and 5,000 calls per day. Enterprise is custom.

Bland's tier math has two clean breakpoints. Build becomes cheaper than Start above 14,950 minutes per month. Scale becomes cheaper than Build above 20,000 minutes. Below those marks, the platform fee can erase the lower minute rate.

Vapi beats Bland Start only while its combined STT, LLM, and TTS providers stay below $0.09 per minute, before telephony. Bland is the stronger budgeting choice when a flat AI rate matters more than swapping providers. Its limits are the platform fees on higher tiers, daily call caps, and separate telephony.

Who should pick Vapi, and who should skip it

Pick Vapi if you have technical ownership of the stack, want to choose each provider, can reconcile multiple invoices, and expect the combined STT, LLM, and TTS rate to remain below your alternative's break-even. Ten concurrent calls should also cover the normal peak, or usage should be high enough to dilute added line costs.

Pick Retell AI if 20 included concurrent calls remove a meaningful fixed cost, or if your chosen Vapi provider stack exceeds $0.065 per minute against Retell's $0.115 example. It is also easier to evaluate when you want a published component calculator inside one platform.

Pick Bland AI if invoice predictability is the priority and $0.14 per minute for bundled LLM, STT, and TTS fits the unit economics. Move to Bland Build only above 14,950 minutes and to Scale only above 20,000, unless the higher concurrency or daily cap is itself worth the platform fee.

Get written pricing before implementation when you need HIPAA, Zero Data Retention, enterprise identity controls, or more than 50,000 monthly minutes. The first two are published Build add-ons; the latter requirements may push the project to Scale. Either way, a five-cent comparison is misleading.

The Monday move

On Monday, pull a representative month of connected minutes, average call length, peak simultaneous calls, and the outcome rate that matters, such as booked appointments. Add the per-minute STT, LLM, TTS, and transport charges from the provider invoices you would use.

Price Vapi as minutes multiplied by the full variable stack, then add extra lines and compliance. Price Retell from the exact calculator configuration, not the bottom of its range. Price Bland at the tier that matches both minutes and peak capacity.

Then run the two break-even tests: provider costs below $0.065 favor Vapi over Retell's $0.115 example, and costs below $0.09 favor Vapi over Bland Start. If compliance is mandatory, confirm the Build add-ons in writing; if volume approaches 50,000 minutes, get a Scale quote before a production build. Finally, cap inbound traffic and set auto-reload deliberately so success cannot turn into an uncontrolled bill.

How much does Vapi cost per minute?

Vapi charges $0.05 per call minute for hosting on Build. STT, LLM, TTS, and transport are additional, so the finished rate is $0.05 plus those provider costs and any fixed add-ons.

Is Vapi AI expensive?

Vapi is inexpensive when 10 concurrent lines are enough and the provider stack stays lean. It becomes expensive at low volume when extra concurrency, the $2,000 HIPAA add-on, or the $1,000 Zero Data Retention add-on must be spread across few minutes.

Do you have to pay for Vapi?

Not for a short prototype inside the 60+ included minutes shown on the Build page. Production calls are usage based, and provider or transport charges can still apply separately.

Does Vapi have a pricing calculator?

Yes. Vapi's live pricing page has a calculator with separate inputs for hosting, transport, STT, LLM, and TTS. Use the result as a configuration estimate, then add concurrency and compliance costs.

Does Vapi offer a student discount?

No public student discount is listed on Vapi's current pricing page or in its terms. The included trial minutes are the available no-cost entry point.

What is Vapi's refund policy?

Vapi does not publish a blanket refund guarantee on its pricing page. Its terms mention pro-rata refunds in limited Vapi-initiated termination and infringement-remedy situations, so ask support before prepaying if refundability matters.

What is the cheapest AI calling-agent option?

There is no universal cheapest option. Vapi wins when its provider stack stays below $0.065 against Retell's $0.115 example or below $0.09 against Bland Start; peak concurrency, telephony, and compliance can reverse both results.

If you want this cost model turned into a guarded production receptionist, see AI customer service systems.