TL;DR: Vapi pricing starts at a platform fee of about $0.05 per minute, but that only covers Vapi's orchestration layer. A real voice agent also pays separately for speech-to-text (~$0.01/min), the language model (~$0.02–$0.20/min), text-to-speech (~$0.04/min) and telephony (~$0.01/min). Add it up and the realistic all-in cost is roughly $0.07–$0.25 per minute, rising past $0.30 with premium voices and large models. There is no mandatory base subscription on pay-as-you-go; an Enterprise plan with HIPAA options is sold as a custom, undisclosed contract. Budget on the all-in number, not the headline.
How Vapi pricing works in 2026
The single most important thing to understand about Vapi pricing is that Vapi sells an orchestration layer, not an all-inclusive voice agent. Its advertised rate — around $0.05 per minute — is the fee for Vapi's own software: the thing that ties together the other pieces and gives you a clean API to control them. The pieces it ties together each carry their own cost, and those pieces are where most of your bill comes from. This is not a hidden fee or a trick; it is how a model-agnostic platform works. But it does mean that anyone budgeting from the headline number alone will be unpleasantly surprised by their first invoice.
A working voice agent is a pipeline. Audio comes in and is transcribed to text by a speech-to-text (STT) engine. That text goes to a large language model (LLM) that decides what to say and whether to call a tool. The response is turned back into speech by a text-to-speech (TTS) voice. And the whole conversation rides over a phone line or web connection provided by a telephony layer. Vapi orchestrates this in real time; the STT, LLM, TTS and telephony providers each bill you separately for their part.
The real per-minute cost, component by component
| Component | Typical cost / minute | What drives it |
|---|---|---|
| Vapi platform fee | ~$0.05 | Vapi's orchestration layer |
| Speech-to-text (STT) | ~$0.01 | Transcription provider and accuracy tier |
| LLM processing | ~$0.02–$0.20 | Model size; token volume per turn |
| Text-to-speech (TTS) | ~$0.04 | Voice provider; premium voices cost more |
| Telephony | ~$0.01 | Carrier, region, phone-number type |
| Realistic all-in | ~$0.07–$0.25+ | Sum of the above; $0.30+ with premium choices |
Component ranges reflect widely reported 2026 figures and Vapi's published pricing structure. Your exact costs depend on the providers you select; verify current rates on Vapi's official pricing page and on each component provider's pricing.
The LLM line is the one that swings your bill the most. A small, efficient model handling simple scripted turns might cost a couple of cents a minute; a large frontier model reasoning over a long context can be ten times that. This is the first place to look when costs feel high, and it is also the easiest lever to pull without hurting the experience for straightforward use cases.
Worked examples: what a call actually costs
Abstract ranges are hard to budget against, so here are three illustrative scenarios. These are estimates built from the component ranges above, not quotes — your numbers will differ — but they show how configuration drives cost.
Example 1: lean appointment-reminder bot
Short, scripted outbound calls with a small LLM, a standard voice and basic telephony might run near the bottom of the range — call it roughly $0.08–$0.12 per minute all-in. At an average call length of 90 seconds, that is on the order of $0.12–$0.18 per call. For a campaign of 10,000 reminder calls, you are looking at roughly $1,200–$1,800 in usage, dominated by the LLM and TTS components rather than Vapi's fee.
Example 2: customer-support agent with tool calls
Inbound support calls are longer, more conversational, and use a more capable model plus a better voice for brand reasons. That pushes you toward the middle-to-upper range — perhaps $0.15–$0.25 per minute all-in. A five-minute support call then costs roughly $0.75–$1.25. The economics still compare favourably with a human agent, but you should model real average handle times rather than assuming best-case.
Example 3: premium consumer concierge
A consumer-facing concierge using a premium voice and a large frontier model for natural, flexible conversation can exceed $0.30 per minute. Here the experience justifies the cost, but it is essential to monitor usage closely, because a small increase in average call length multiplies across volume.
Comparing platforms on real cost? See our Vapi vs Retell breakdown, or read the full Vapi review in our voice AI agents coverage.
The Enterprise plan
Beyond pay-as-you-go, Vapi offers an Enterprise plan that adds unlimited concurrency, 24/7 support and HIPAA-compliant configurations for regulated use cases such as healthcare. Its pricing is a custom contract and is not publicly disclosed, so we will not invent a figure. If you are running high call volumes, need guaranteed concurrency for spikes, or must meet compliance requirements, this is the tier to discuss with Vapi's sales team. For everyone else, pay-as-you-go with no base subscription is the friendlier starting point, especially for prototyping.
Hidden costs and gotchas
- Model overkill: using a large frontier model for simple flows is the most common source of avoidable cost. Match the model to the task.
- Wasted minutes: poor interruption handling and long silences inflate call duration. Tuning turn-taking saves real money at volume.
- Premium voices: a more lifelike voice can double your TTS line. Decide whether the use case justifies it.
- Telephony quirks: international calls, toll-free numbers and certain regions cost more than the baseline.
- Failed and abandoned calls: you may still incur some component costs on calls that do not complete. Track cost per completed outcome.
How to estimate and control your Vapi bill
Start by building one narrow agent end to end and running real test traffic through it. That single exercise reveals your actual per-minute cost far better than any table, because it captures your specific model, voice, transcription and telephony choices and your real average call length. From there, the levers are straightforward: right-size the LLM, choose an efficient voice, keep prompts and context lean, and design conversations to be concise. Instrument your calls so you can see cost per completed call over time, and watch for drift as traffic patterns change. Vapi provides trial credits precisely so you can do this measurement before committing to volume — use them.
Finally, model at scale before you scale. A cost that feels trivial per call becomes a real budget line at 100,000 calls a month, and the components that dominate at low volume (often the LLM and TTS) are exactly the ones to negotiate or optimise as you grow. The teams that are happiest with Vapi pricing are the ones that treated cost as an engineering parameter from day one, not an afterthought.
Vapi pricing vs alternatives
Vapi is not the only developer-first voice platform, and a sensible budgeting exercise compares it against peers on an all-in basis. Retell AI advertises a $0.07/minute conversation rate that, all-in, typically lands around $0.13–$0.31/minute — broadly overlapping with Vapi once components are included. Voice quality specialists like ElevenLabs are often used as the TTS layer inside platforms like Vapi, which means their cost may already be part of your TTS line rather than a separate choice. The honest conclusion is the same one we reach in our Vapi vs Retell comparison: do not choose on headline rate. Build the same agent on each candidate, measure cost per completed call on real traffic, and decide from data.
When does a Vapi voice agent pay for itself?
Cost per minute only matters relative to what the agent replaces or enables. Against a human agent — whose fully loaded cost per minute of talk time is far higher than even a premium voice agent — the economics usually favour automation for high-volume, repetitive calls such as reminders, confirmations, qualification and tier-one support. The break-even is not really about the per-minute rate at all; it is about whether the agent completes the outcome reliably enough to avoid expensive human follow-up. An agent that costs $0.15 a minute but resolves the call beats one that costs $0.08 a minute but routinely escalates, because the escalation is where the real money goes. This is why we keep returning to cost per completed outcome as the metric that matters. When you evaluate Vapi pricing, weigh it against the loaded cost of the human work it offsets and the value of the outcomes it delivers, not against an abstract ideal of cheapness. Framed that way, the realistic $0.07–$0.25 all-in range is affordable for a wide span of use cases, and the platform fee itself — the part that is uniquely Vapi — is rarely the deciding cost.
Frequently Asked Questions
How much does Vapi really cost per minute in 2026?
Vapi's platform fee starts around $0.05 per minute, but that covers only its orchestration layer. Once you add transcription, the language model, text-to-speech and telephony, a realistic all-in cost lands between roughly $0.07 and $0.25 per minute, and can exceed $0.30 with premium voices and larger models.
Is there a base subscription for Vapi?
No. Vapi's pay-as-you-go model has no mandatory base subscription — you pay for usage. There is a separate Enterprise plan with unlimited concurrency, 24/7 support and HIPAA-compliant options, sold as a custom contract whose price is not publicly disclosed.
Why is Vapi more expensive than its advertised rate?
Because the advertised $0.05/minute is only Vapi's orchestration fee. A working voice agent also pays separately for speech-to-text, the large language model, text-to-speech and telephony. Those components, not Vapi itself, are where most of the per-minute cost comes from, and they vary with the providers you choose.
How can I reduce my Vapi costs?
The biggest lever is model choice: a smaller, cheaper LLM for simple scripted turns can cut cost dramatically versus a large frontier model. Choosing efficient transcription and a reasonably priced voice, keeping calls concise, and handling interruptions well to avoid wasted minutes also help. Measure cost per completed call, not per minute in isolation.
Is Vapi cheaper than Retell?
On headline rates Vapi (from ~$0.05/min) looks cheaper than Retell (from $0.07/min), but both are only one layer of the total. All-in, Vapi typically runs ~$0.07–$0.25/min and Retell ~$0.13–$0.31/min depending on components. The cheaper option depends on your configuration; see our Vapi vs Retell comparison for details.
Forecasting voice AI costs for a contact centre? Talk to our editors →