Vapi vs Retell AI: Which Voice Agent Platform Actually Costs Less?

Two platforms turn up in almost every voice agent shortlist, and most comparisons between them read like feature checklists written by someone who has deployed neither. Both build phone agents. Both handle inbound and outbound. Both connect to Twilio. Stop at that level and the choice looks arbitrary.

It is not arbitrary. Vapi and Retell AI differ on one structural decision, and once you can see it, most of the other differences fall out of it as consequences. This is a comparison of that decision, using the current published numbers from both, ending in a rule you can actually apply.

Every figure here was taken from the two vendors on 13 August 2026. Voice AI pricing moves faster than almost any category in software. Check both pages before you commit a budget to either.

The One Difference That Decides It

A voice agent call is four services stitched together: speech-to-text, a language model, text-to-speech, and telephony. Neither Vapi nor Retell builds any of those. Both orchestrate them. What separates the two platforms is what happens to the bill for the four.

Retell prices them itself. Its pricing page carries a full rate card: GPT 5.1 at $0.04 a minute, Claude 5 Sonnet at $0.08, ElevenLabs voices at $0.040 against $0.015 for the platform voices, Twilio telephony at $0.015. You need no account with any of those providers. You never see their meters. You pay Retell a per-minute number and Retell settles up behind the scenes.

Vapi passes them through. Its cost routing documentation is unusually blunt about the mechanism. If you have a provider key on file, the request goes out with your key and the charge lands in your account with that provider, with nothing billed by Vapi at all. If you have no key on file, Vapi makes the request with its own and bills you at cost. The documentation states plainly that you are never charged any margin on provider fees.

That is the entire comparison in miniature. Retell sells you a finished per-minute price. Vapi sells you orchestration and lets the pipeline bill you directly.

Everything below is downstream of that.

What Vapi Actually Charges

Vapi runs two plans. Build is self-serve and usage-based. Scale is an annual contract with a fixed platform fee, committed volume and volume-based per-minute rates, quoted by their sales team rather than published.

Here is the Build plan, in full, from Vapi's pricing page:

Two of those rows deserve more attention than they usually get.

Call history is kept for 14 days. If you need recordings, transcripts or call metadata beyond a fortnight, and most businesses with any quality-assurance process do, you have to export them yourself into your own storage. That is an engineering job nobody budgets for because nobody reads the retention row.

HIPAA has a published price. $2,000 a month, available on the self-serve plan without an enterprise contract. That is unusual and genuinely useful if you are building for a clinic and want to know the number before booking a sales call. It is also a large number relative to a small deployment, so it tends to force the decision either way rather than leaving it open.

What Retell Actually Charges

Retell runs pay-as-you-go and Enterprise. Pay-as-you-go starts at $0 with $10 in free credits, no contract and, in Retell's framing, no feature gating on the core platform.

The rate card, layer by layer:

The headline band Retell advertises is $0.07 to $0.31 a minute, which is wide because it is the range you land in after choosing a model and a voice, not a rate anyone actually pays.

Retell also publishes its billing mechanics, which is worth something on its own. Calls are tracked to the nearest second with no per-call rounding. Failed calls are not billed. When a call transfers to a human, the AI fee stops and only the telephony fee continues. And billing runs through silence and hold, because the speech-to-text engine stays active and listening the whole time. Vapi's pricing page does not spell out equivalent rules. If you have ever tried to reconcile a voice bill, you will know how much that documentation is worth.

I broke the Retell card down further, including the add-ons that catch people out, in the full Retell AI review.

The Platform Fee Is a Red Herring

Put the two orchestration fees side by side and the difference evaporates. Vapi charges $0.05 a minute. Retell charges $0.055. Half a cent apart. Over ten thousand minutes a month, that gap is fifty dollars, which is less than the noise in either bill.

So the platform fee is not the decision. The pipeline is.

Take a realistic workload: 10,000 minutes a month, a mid-tier model, a standard voice, on a Twilio number.

On Retell, the card gives you the whole answer. $0.055 for infrastructure, $0.015 for a platform voice, $0.015 for telephony and $0.04 for a GPT 5.1-class model comes to $0.125 a minute. That is $1,250 a month, plus $2 for the phone number. Swap Retell's telephony for your own SIP trunk and it drops to $0.11 a minute, which is exactly the figure Retell's own calculator produces for that stack: $1,100.

On Vapi, you know one number and have to supply the rest. $0.05 times 10,000 minutes is $500 in platform fees. Then the pipeline bills you separately, and nobody can quote that figure for you, because it depends entirely on which speech-to-text engine, model and voice you pick and what you pay for them.

Which gives you the actual question, and it is not the one people ask:

Can you assemble the speech-to-text, model, voice and telephony for less than 7.5 cents a minute? Below that, Vapi is cheaper. Above it, Retell is.

That is a question about you, not about the platforms. It depends on whether you already run accounts with these providers, whether you have negotiated anything on volume, and whether you have the appetite to tune a pipeline rather than pick from a dropdown. Someone already spending real money with OpenAI and a voice provider will beat 7.5 cents comfortably. Someone starting from zero, buying at list price and learning which combinations sound acceptable, very often will not.

If you want to run your own version of that arithmetic before committing, the voice AI cost per minute calculator lets you set each layer independently, and the AI voice agent cost simulator models it against a monthly call volume rather than a per-minute rate.

Concurrency: Where Retell Wins Outright

Concurrency is how many calls the agent can hold at once, and it is the line item that quietly decides the bill for anyone running campaigns rather than a single reception line.

Retell includes 20 concurrent calls and charges $8 per concurrent call per month above that. Vapi includes 10 lines and charges $10 per line per month above that. Both are cheap in isolation. They are not the same at scale.

Retell is cheaper at every point above ten, and the gap widens as you grow. At a hundred concurrent lines it is $260 a month, or a bit over three thousand a year, for capacity alone.

This matters most for outbound. An inbound receptionist rarely needs more than a handful of simultaneous calls. A dialling campaign needs as many as you can afford, and the arithmetic above is a real constraint on how fast you can work a list. If that is your use case, the mechanics are covered in more depth in the guide to running outbound AI calling campaigns.

Retell also charges an extra $0.005 per dial for batch calling and $0.10 per outbound call for branded caller ID. Two thousand branded dials is $200 on top of everything else, so budget it deliberately rather than switching it on because it sounds good.

Compliance, Security and Data

This is where the two platforms diverge on philosophy rather than price.

Vapi publishes prices for compliance add-ons but gates the security controls. HIPAA is $2,000 a month and zero data retention is $1,000 a month, both available on the self-serve Build plan. But SSO, role-based access control and SOC 2 coverage are Scale-only, which means an annual contract and a sales process. You can therefore end up in the odd position of being HIPAA-covered while your team still shares a login.

Retell now gives away the expensive part of compliance. Its compliance documentation offers the BAA and the DPA, including EU Standard Contractual Clauses, for self-signing at no additional fee on the pay-as-you-go plan, with per-agent data retention configurable from one day to two years and per-agent PII controls. What still sits on Enterprise is custom SSO, role-based access control, custom MSA terms, dedicated servers and 24/7 support. The practical effect is that the same signature costs nothing on Retell and $2,000 a month on Vapi. In exchange, the pay-as-you-go plan gives full platform access with no feature gating: every agent capability, the API, webhooks, simulation testing, analytics and transcripts, from a $0 starting point.

The practical read is this. If you are a small team building something regulated and you want a price today rather than a meeting, Vapi will give you one. If you are a small team building something unregulated and you want every product feature without a plan upgrade, Retell will give you that. If you are an enterprise, both end in a sales conversation and the comparison becomes about the quote, not the card.

Worth noting on retention specifically: Vapi's 14-day call history on Build is a hard operational constraint, while Retell offers custom data retention only on Enterprise. Neither self-serve plan is a records system. Plan on exporting to your own storage in both cases.

Control, and What You Give Up for It

Vapi's pass-through model buys you provider freedom. Any speech-to-text engine, any model, any voice vendor, using your own keys, with your own negotiated rates and your own rate limits. If a better model ships next Tuesday, you can be on it that afternoon without waiting for a platform to add it to a dropdown.

Retell narrows that deliberately. Six text-to-speech providers, a fixed list of models across the GPT, Claude and Gemini families, and Twilio or Telnyx for telephony. It is not a closed box, though, and the comparisons that call it one are being lazy: Retell supports a custom LLM endpoint and custom telephony via your own SIP trunk, which covers the two places where teams most often need to escape a managed stack.

So the honest framing is not open versus closed. It is how much of the stack you want to be responsible for. Vapi hands you the pipeline and the bill for it. Retell hands you a price and keeps the pipeline. Both are defensible, and the wrong one for you will feel wrong for exactly the reason you would predict.

There is a hidden cost on the Vapi side that rarely appears in comparisons. Managing four provider accounts means four sets of API keys, four billing relationships, four rate limits to monitor and four places to look when latency spikes. That is fine if you are a developer or you have one on the team. It is a genuine burden if you are an operator who wanted a phone agent.

Which One Fits Which Job

Choose Vapi if you are a developer or an agency with engineering capacity, you already have provider accounts and want to keep your own rates, you care about picking exactly which model and voice run in the pipeline, or you expect to swap components as the model landscape moves.

Choose Retell if you want one predictable per-minute number you can quote to a client, you are running enough concurrency that the $8 versus $10 per line difference compounds, you value published billing mechanics like per-second tracking and no charge on failed calls, or you want the full product surface without a plan upgrade while spending very little.

Either works if you are building a single inbound receptionist doing a few hundred minutes a month. At that volume the total bill is small enough that the decision does not deserve a week of research, and the difference between the two will be under twenty dollars. Pick the one whose builder you find more pleasant and get the thing live. If that is the project, the AI receptionist build guide covers the parts that actually determine whether it works, which are the prompt and the call flow rather than the platform.

Whichever you pick, the telephony layer is where most first deployments break rather than the agent itself. The walkthrough on connecting a Twilio number to a voice agent covers the configuration both platforms assume you already understand, and Twilio's own Programmable Voice documentation is the reference for the underlying behaviour.

The Decision Rule

If you cannot decide, use this and move on.

Estimate what your speech-to-text, model, voice and telephony would cost per minute if you bought them yourself. If that number is comfortably under 7.5 cents, Vapi is cheaper and the pass-through model is working for you. If it is above 7.5 cents, or if you cannot estimate it with any confidence, Retell's card is both cheaper and easier to plan around.

Then check concurrency. If you need more than twenty simultaneous calls, Retell's included allowance and lower per-line rate will pull the total further in its favour regardless of the first answer.

Then check compliance. Both platforms will sign a BAA without an enterprise contract, but Retell does it for nothing where Vapi charges $2,000 a month, so this check no longer points at Vapi the way it did a year ago. The one thing neither fixes is processing inside the EU, which is covered in the alternatives worth considering.

Three checks, in that order. Most teams reach a clear answer by the second.

If you would rather not run the comparison at all, I build and deploy voice agents on both platforms as a service. You can see the current packages on Fiverr or hire me directly through Upwork, and the step-by-step Retell build guide is there if you would rather do it yourself.

Frequently Asked Questions

Is Vapi cheaper than Retell AI?

It depends entirely on what your pipeline costs. The two platform fees are nearly identical at $0.05 and $0.055 a minute. Everything else on Vapi is billed at cost or by your own providers, while Retell charges its own published per-minute rates for the model, voice and telephony. If you can assemble the pipeline for under roughly 7.5 cents a minute, Vapi comes out cheaper. If you cannot, or you buy everything at list price, Retell usually does.

Does Vapi add a markup to model and voice costs?

No. Vapi's cost routing documentation states that you are never charged any margin on provider fees. If your own provider key is on file, the charge appears in your account with that provider and Vapi bills nothing for it. If no key is on file, Vapi makes the request with its own key and passes the cost through at cost.

Which platform is better for outbound calling campaigns?

Retell, on cost. It includes 20 concurrent calls against Vapi's 10, and charges $8 per additional concurrent call per month against Vapi's $10 per line. At fifty concurrent calls that is $240 a month versus $400. Budget separately for Retell's batch calling at $0.005 per dial and branded caller ID at $0.10 per outbound call, both of which add up quickly at campaign volume.

Can I use my own phone number and SIP trunk on both?

Yes. Retell supports custom telephony and charges nothing for calls placed through your own SIP trunk, against $0.015 a minute for its Twilio or Telnyx numbers. Vapi bills telephony wherever the number was provisioned, so an imported Twilio or Vonage number is charged in that account rather than by Vapi. Bringing your own trunk is usually the single largest saving available on either platform.

How long do the two platforms keep call recordings?

Vapi's Build plan retains call history for 14 days and chat history for 30 days, with custom retention on the Scale plan. Retell offers custom data retention on its Enterprise plan. Neither self-serve plan should be treated as a records system, so if you need transcripts or recordings for quality assurance or compliance, plan on exporting them into your own storage from day one.

Do I need to be a developer to use either one?

Not strictly, but Vapi assumes more. Its pass-through model means managing provider accounts, API keys, billing relationships and rate limits across up to four vendors, which is straightforward for a developer and a real burden for an operator. Retell's managed rate card removes that entirely: you pick a model and a voice from a list and receive one bill. If you have no engineering support, that difference matters more than the per-minute figures.