Vapi AI Pricing: What the Per-Minute Model Actually Costs

Vapi AI pricing is usage-based, with platform fees charged per minute. Then there's a separate cost for each speech-to-text, language model, and text-to-speech provider that you connect to.

This is the one detail most pricing pages skip. It is also the reason the number advertised on the homepage rarely matches what appears on the invoice.

Let's break down what Vapi actually costs beyond its per-minute rate. That means counting the engineering hours it takes to turn a voice API into an agent that holds up in production. Because that’s an entirely different number that many businesses fail to budget for.

How Vapi's Pricing Model Works

Vapi charges its own per-minute platform fee for running the call itself. It then passes through the separate costs of every speech-to-text, language model, and text-to-speech provider you connect.

This goes to show you will still be paying a stack of charges rather than a single number.

The Vapi platform fee

Vapi's self-serve Build plan lists a platform fee of $0.05 per minute for calls and $0.005 per message for SMS and chat. It gives you 60+ minutes and 10 concurrent calls. Additional lines cost $10 per month each.

Now, this covers just the platform fee, which includes Vapi's side of running the call. It keeps the call connected and manages the handoff between the speech and language models in real time. It does not cover the models that actually run the call, such as speech recognition or language models.

What you pay on top of it

The model costs are billed separately at the provider's rate unless you bring your own API key, in which case Vapi doesn't mark them up at all.

A call running a premium language model and a premium voice provider costs significantly more per minute than the same call running a cheaper stack, even though Vapi's own platform fee stays flat at $0.05 either way.

For example, Claude 3.7 Sonnet costs $0.09/1M tokens, whereas the GPT-4o Realtime Preview costs $1.14/1M tokens.

On top of that, you have text to speech layer for which you’re charged extra per minute.

Since you cannot plan the number of minutes for calls beforehand, you cannot budget for Vapi, which is also why you cannot compare Vapi to a flat-fee product with a single number.

The enterprise path

For larger enterprises, Vapi also offers a Scale plan that requires an annual contract with a fixed platform fee and a committed call volume. This plan includes SOC 2, HIPAA, SSO, and a dedicated account team. 

Pricing on both plans is subject to change, so treat every figure here as reported and confirm current rates on Vapi's official pricing page before you budget against them.

How Much Does Vapi AI Cost Per Minute?

Vapi charges a flat $0.05 platform fee per minute. On top of that, you pay for whatever speech-to-text, language model, and text-to-speech providers you choose. 

Add it all up, and independent testing puts the real cost between $0.07 and just over $1.00 per minute.

What drives the cost:

• Platform fee: Fixed at $0.05 per minute, no matter what you build
• Speech-to-text: $0.008 to $0.017 per minute, depending on the provider
• Language model: Under a cent to $0.32 per minute, depending on which model you pick
• Text-to-speech: A fraction of a cent to $0.65 per minute for premium voices

What that means in practice:

• Cheapest stack + platform fee: around $0.07 per minute
• Most expensive stack + platform fee: can pass $1.00 per minute

These numbers come from Softailed's independent testing, not from Vapi's own published figures. A separate review from Lindy found a narrower range, $0.07 to $0.25 or more, using the same variables. Both are third-party estimates. Treat them as a starting point while checking your own stack's cost before committing to volume.

One thing the per-minute price doesn't include: build time

A developer on Reddit documented 760 calls and over 100 hours of work building a Vapi agent. Their finding: the AI itself was roughly 20% of the effort. The rest went into error handling and the automation workflows needed to feed call outcomes back into a CRM.

This post discusses building an after-hours agent to help customers look up specific items in a Shopify store with 4,500+ products. It needed a separate workflow to pull the right products from a disorganized catalog and hand the AI a clean answer, all within 800 milliseconds, before the call went silent. That one piece took days to get stable.

The replies say the same thing. One developer switched platforms because when a Vapi call broke, the logs didn't show what went wrong. Another posted a live bug: a workflow that worked on a test URL stopped firing the moment it went to production, with no logs to explain why.

This kind of build time isn't unique to Vapi. It shows up across most modular voice platforms. Also, the reason why you’d budget for it before comparing Vapi's advertised rate against a fully managed alternative.

Vapi's Pricing Plans

Vapi runs two plans:
I. Build - a self-serve, usage-based plan for individual developers and small teams

II. Scale -  an annual enterprise contract with custom volume pricing.

Here’s more in detail.

Vapi Build (self-serve, usage-based) plan

Build is where most people start. New accounts get a one-time $10 trial credit rather than an ongoing free tier, and once that credit runs out, every call is billed at the standard per-minute rates.

Other inclusions:

• 10 concurrent calls; extra lines cost $10 per line, per month
• Call history is stored for 14 days; chat history for 30 days
• Does not include SSO, RBAC, or SOC 2

Vapi Scale (enterprise, annual contract) plan

Scale is the enterprise path that runs on a fixed platform fee and a committed call volume.

You negotiate three things: 

• A fixed platform fee
• A committed call volume
• The per-minute rate that volume buys you. 

Vapi doesn't publish the Scale rate, so the number you pay depends on what you commit to.

Non-negotiated:

• The HIPAA add-on ($2,000/month)
• Zero Data Retention ($1,000/month) costs the same on Scale as they do on Build.

Other inclusions:

• SOC 2, HIPAA, and PCI compliance
• SSO, RBAC, and a dedicated account team

Good to know: model provider costs
On the Build plan, Vapi says model costs (STT, LLM, TTS) are billed to you at cost. They do come free if you bring your own API key.

For Scale, those same cells in their table are blank. So we don't know if enterprise customers still pay those model costs separately, or if they're folded into the contract. But for an enterprise weighing Vapi against a fully managed alternative, the real comparison point is the Scale plan since that's the tier built to compete with a deployed voice agent on cost and support.

Is Vapi Worth It, or Should You Deploy a Voice AI Agent Instead?

Vapi is worth its modular pricing if your team has the engineers to assemble, test, and maintain a voice pipeline. If you don't have those engineers, a low per-minute rate buys you very little.

VAPI average cost per call | Source: YouTube

You are essentially still paying for the build, just through engineering salaries rather than usage fees.

Vapi fits teams that want to choose their own speech, language, and voice providers and are ready to own error handling and the automation that turns a finished call into a CRM update or a follow-up message. That’s a real control, but it comes at the cost of the engineering time required to build and maintain it.

HappyRobot as Vapi Alternative: AI Voice Agent 

HappyRobot deploys AI workers built on a proprietary voice stack that's already production-ready, supports 30+ languages, and goes live under a contract instead of a per-minute-plus-components bill you assemble yourself.

There's no separate STT, LLM, and TTS invoice to reconcile every month, and no webhook layer to maintain when a provider changes its API.

Forward Deployed Engineers handle the buildout directly. Once the agent is live, every inbound call connects as soon as it arrives, with a 100% answer rate and no hold queue.

Vapi gives you control over every layer of the voice pipeline. HappyRobot gives you the agent already built and running.

Vapi figures reflect the company's published rates as of the writing and are subject to change. Please confirm current pricing directly with Vapi before budgeting.

HappyRobot figures reflect its published deployment and contract model. For a full side-by-side on architecture and workflow depth, see Vapi vs. HappyRobot.