Retell AI vs Vapi (2026): Which Is Cheaper at 10,000 Minutes?

At 10,000 minutes a month our Retell AI support stack costs $1,144 and Vapi costs $819 to $1,290 on its calculator defaults, so Vapi is cheaper only with a small model; Retell is the easier pick for an ops team, with a visual builder and 20 concurrent calls included. Most pages ranking for this search have a stake in it: Retell wrote one that shows Vapi at 2.4 stars, and most others come from rivals such as Bland, Synthflow and Famulor. Macha doesn't sell a voice platform and has no stake in either. Our Retell AI guide and Vapi guide cover each product on its own; this page only compares them.

How do Retell AI and Vapi compare at a glance?

Prices rechecked September 24, 2026 and ratings checked September 19, 2026 on retellai.com/pricing, vapi.ai/pricing, Trustpilot and Gartner Peer Insights. Customers are from each vendor's customers page (Retell, Vapi); the results those pages claim are the vendors' own.

How do Retell AI and Vapi bill a call?

Both platforms bill the same four things on a call: listening (speech-to-text), thinking (the model), speaking (text-to-speech) and the phone line. They differ in who sets the price of each.

Retell sets all four. Its voice infrastructure is $0.055 a minute, which includes speech-to-text, its standard voices are $0.015, telephony on a Retell number is $0.015, and the model runs from $0.0016 (GPT 5 nano) to $0.32 (GPT 6 Astra) a minute. Those are Retell's prices, so when Retell pays ElevenLabs for a voice, you pay Retell's $0.040 a minute for "Elevenlabs Voices", not ElevenLabs' rate.

Vapi sets one. Its hosting fee is $0.05 a minute, and its pricing FAQ says the transcriber, model, voice and transport are each billed at the provider's listed price with no markup. On Vapi's calculator the same ElevenLabs voice costs $0.0146 to $0.0238 a minute. You can also plug in your own provider keys, in which case the provider bills you and Vapi only charges hosting.

What each model rewards. Retell earns a margin on every component, so it gains when you pick a heavier model or a premium voice; in exchange it gives you one invoice and one number to budget. Vapi earns the same $0.05 whatever model you run, so it has no reason to steer you to an expensive one; its extra revenue comes from minutes and from the packages that buy support and capacity. On both, a longer call earns the vendor more whether or not it solved anything.

What do Retell AI and Vapi cost at 10,000 minutes a month?

We set both public calculators to 10,000 minutes a month, about 2,500 four-minute calls on an inbound support line.

On Retell we chose GPT 5.4 mini (a mid-priced model that handles tool calls), Retell's platform voices and Retell's Twilio/Telnyx telephony. The calculator returned $0.109 a minute and $1,090 a month.

The calculator leaves out two things a support line needs: the knowledge-base add-on ($0.005 a minute, which didn't register in the calculator when our Retell pricing research tested it; the first 10 knowledge bases are free, then $8 each a month) and phone numbers at $2 a month each. With both, Retell comes to $0.114 a minute: 10,000 × $0.114 = $1,140, plus $4 for two numbers, or $1,144 a month.

On Vapi we clicked 10k and kept its defaults: Deepgram, OpenAI and ElevenLabs, on Vapi telephony. The estimate was $819 to $1,290, with no package.

Vapi's range comes from the model line ($77 to $452 across OpenAI's models) and the voice ($146 to $238). The estimate also shows 4 concurrent calls, which a 2,500-call line will outgrow at peaks. Core, at $29 a month, raises that to 10, so a realistic Vapi month is $848 to $1,319.

The same arithmetic at three volumes, with each platform's cheapest option for the concurrency (our assumption: peaks under 4 calls at 1,000 minutes, 8 at 10,000, 25 at 50,000):

Retell sits inside Vapi's range at every volume. Vapi is cheaper if you run a small model and cheaper still with your own negotiated provider contracts; it's dearer if you run its top OpenAI model through ElevenLabs.

Two costs change the picture more than the per-minute rate:

• HIPAA. Vapi charges $2,000 a month per organization. At 1,000 minutes that turns an $82 to $129 bill into $2,082 to $2,129. Retell's plan comparison table ticks HIPAA/BAA on pay as you go.
• Support with a response time. Vapi's usage-only tier comes with "Peer to peer support via Discord community" and "Email support without SLA"; Core adds a 2-business-day SLA and Pro (10% of hosting, $999 minimum) a 1-business-day SLA. Retell's pay-as-you-go plan lists community and email support, and its 24/7 portal is an Enterprise item.

Chat agents. Retell prices chat per AI message by model, from $0.0005 (GPT 5 nano) to $0.10 (GPT 6 Astra); GPT 5.4 mini is $0.0075. A month of 3,000 web chats with 6 AI replies each is 18,000 messages: $9 on GPT 5 nano, $135 on GPT 5.4 mini. Vapi has a Chat API, but its pricing page lists voice minutes only, and its chat docs require a card on file and a completed credit purchase for new pay-as-you-go organizations. We couldn't price a Vapi chat month from its public pages.

Who can build an agent on Retell AI and on Vapi?

Retell gives you two ways to build. Prompt agents (single or multi-prompt) suit open conversations, and conversation flow agents lay the call out as nodes on a canvas: conversation, function, call transfer, press digit, SMS, logic split and end. An ops lead can edit a flow in the browser without touching code. Retell's docs also document integration tools for HubSpot, Salesforce, Zendesk and Cal.com, MCP tools, and a code node that runs JavaScript mid-call.

Vapi is configuration first. An agent is an assistant (prompt, model, voice, tools), and multi-step behavior comes from Squads, where specialized assistants hand a call to each other. Vapi retired its visual Workflows builder on August 18, 2026; its migration guide says existing workflows stopped running the next day and tells builders to move to assistants or Squads. That leaves Vapi with no drag-and-drop canvas, which suits teams who keep agent config in version control and review changes like code.

The practical split: if the person who owns the agent isn't an engineer, Retell's builder is the reason to choose it. If engineers own it and want to swap a transcriber or model per assistant, Vapi exposes more of the pipeline.

How do transfers, telephony and compliance differ?

Call transfer is where a support agent earns its keep, because the hard calls end with a human.

Retell's transfer tool offers cold transfer, warm transfer (the agent briefs the human first) and agentic warm transfer, where a second AI agent talks to the human and decides whether to bridge the caller. Transfers can go to a phone number or a SIP URI. Its docs note one carrier difference: Retell's Twilio numbers can show the caller's own number on both warm and cold transfers, while Retell's Telnyx numbers can only on cold ones.

Vapi's transfer tool has blind transfer, blind transfer with a summary in a SIP header, five warm-transfer modes and an experimental assistant-based warm transfer. Its warm transfer docs add a carrier limit, which Retell's comparison page leans on.

So if your numbers live on Telnyx, Vonage or a SIP trunk and you need a warm handoff on Vapi, test the assistant-based mode or plan on Twilio.

On telephony more broadly, both bring your own carrier. Vapi's docs include SIP trunk guides for Twilio, Telnyx, DIDWW, didlogic, Zadarma, Plivo and Amazon Chime, and its free numbers are US-only. Retell sells numbers at $2 a month (toll-free and international calls cost more per minute) and imports custom SIP numbers at no monthly charge.

On compliance, Retell's homepage says it is "Fully compliant with HIPAA, SOC2 Type II, GDPR and ISO 27001." Vapi's homepage lists "SOC 2, HIPAA, and PCI compliant," with HIPAA-eligible handling sold as the $2,000 add-on. Check the report and the BAA with either before you send health data.

Is Retell's own Retell vs Vapi page accurate?

Retell's Retell vs Vapi page says its data is "as of 05/2026". We checked its main claims against both vendors' live pages and docs on September 19, 2026.

The page is right about the thing that matters most to an ops buyer: Retell is easier to launch without engineers. Its price and concurrency claims don't match Retell's own pricing page, so budget from that page.

Which has lower latency, Retell AI or Vapi?

Both vendors publish a headline number and neither publishes a shared test. Retell's homepage says "~600ms latency" and credits unnamed "independent benchmarks." Vapi's homepage says "<500ms average latency." Vapi's latency docs add that its displayed total "excludes endpointing and transport time," the pause while the system decides you've stopped talking and the network hop, which callers feel most.

Reviewers report numbers well above both headlines. A Vapi reviewer on Trustpilot, David Hartmann (UAE, March 25, 2026), described "a 3-5 second pause after each turn," and a Famulor comparison puts typical Vapi latency at 1,000 to 1,500ms; Famulor sells a competing platform. Latency on both depends mostly on the model, voice and endpointing settings you choose. Run the same script on each platform with your real carrier before you trust either number.

What do users say about Retell AI and Vapi?

Checked September 19, 2026. The Trustpilot gap is partly method: Retell asks its customers for reviews and Vapi doesn't, so Vapi's 15 skew to people with a complaint. Still, the complaints differ in kind.

Retell's one-star reviews are about support and edge cases. Leshawn Clarke (UK, December 22, 2025) wrote that "they have an AI that replies to you instead" and that Retell points users to Discord. Abdulselam Almnini (Turkey, March 11, 2026) hit a country block on outbound calls to Turkey and called it "Country restrictions with zero transparency or support." Retell's Trustpilot page notes it "Hasn't replied to negative reviews."

Vapi's are about quality and price. Samuel Dane (US, March 28, 2026) said the agent "lags a lot and does not act according to the prompt." Sam Hardwick (UAE, April 2, 2026) wrote that Vapi charges "$1000 for a HIPPA compliant voice receptionist"; the pricing page now says $2,000.

On the positive side, a Retell reviewer, Christian Landsteiner (February 10, 2026, invited), wrote that "Retell really stands out compared to other providers" on "reliability and uptime." Vapi's customers page leans on scale: it says Instawork runs "1M+ minutes of voice screening per month on Vapi."

Should you choose Retell AI, Vapi or neither?

Choose Retell AI if the agent will be built and edited by an ops or support lead rather than an engineer, you want one invoice with telephony and 20 concurrent calls included, or you need HIPAA at low volume without a $2,000 monthly add-on. Budget from its pricing page, not its comparison page.

Choose Vapi if engineers own the agent, you want to pick every provider and pay their list price with no markup, or you already hold enterprise contracts with OpenAI, Deepgram or ElevenLabs and want to use your own keys. Buy Core from day one for the concurrency, and budget for Pro if you need a support SLA.

Consider neither if you want a single all-in rate per minute with the model included (Bland sells that), you want minutes bundled into a plan at a flat $0.08 each (ElevenLabs Agents), or you'd rather a vendor build and run the agent for you. Our guide to conversational AI platforms covers the managed options.

And if most of your support volume is email and chat tickets rather than calls, neither is the right tool. Macha is an AI agent layer on the help desk you already run (Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom): agents read and reply inside the ticket, billing is per ticket from $299 a month for 750 tickets, and setup and monitoring by the Macha team are included. It doesn't answer phone calls. Our guide to AI agents for customer service covers where voice and ticket agents each fit, and the tiers are on our pricing page.

How we researched this

On September 19, 2026 (pricing pages rechecked September 24) we read both pricing pages with a scripted browser, ran both public calculators at 10,000 minutes with the settings named above, and read each vendor's docs on transfers, knowledge bases, SIP trunking, chat billing, latency and (for Vapi) the Workflows retirement. We checked Retell's comparison page claim by claim against those pages, read both Trustpilot profiles including Retell's one-star reviews, and read Gartner Peer Insights, Capterra and G2 in Chrome (Retell 4.8 from 2,638; Vapi 3.0 from 3, checked September 19, 2026). We did not build an agent or open an account on either platform, so every cost here is list-price arithmetic from the calculators and pricing tables, not a bill we paid.

Frequently asked questions

Is Retell AI better than Vapi? For a team without dedicated voice engineers, Retell is easier: a visual flow builder, telephony and 20 concurrent calls included, and far stronger Trustpilot reviews (4.9 from 814 against 2.4 from 15). For engineers who want to choose and pay for each provider directly, Vapi gives more control.

Which is cheaper, Retell AI or Vapi? At 10,000 minutes a month our Retell support stack costs $1,144, and Vapi costs $819 to $1,290 on its calculator defaults, or $848 to $1,319 with the Core package for 10 concurrent calls. Vapi is cheaper with a small model; Retell is cheaper than Vapi's top-model configuration and far cheaper if you need HIPAA at low volume.

Is Vapi AI expensive? Its $0.05 hosting fee is the smallest line on the bill. With default providers a minute costs about $0.08 to $0.13, a support SLA starts with the $29 Core package, and HIPAA is $2,000 a month per organization.

Is Retell AI legit? Yes. Retell is a Y Combinator-backed company with published pricing, public docs and named customers such as Medical Data Systems and Anker. Its Trustpilot profile shows 4.9 from 814 reviews; the profile is claimed and Retell invites customers to review, which lifts the score.

Who are Retell AI's main competitors? Vapi, Bland, ElevenLabs Agents and Synthflow are the platforms most often compared with it; Retell's own site publishes comparison pages against rivals including Bland, Vapi, Synthflow and Thoughtly.

What are Vapi's main competitors? Retell, Bland and ElevenLabs Agents among hosted platforms, and the open-source frameworks LiveKit and Pipecat for teams that want to host their own stack.

How do Retell AI and Vapi compare on latency? Retell advertises about 600ms and Vapi under 500ms, but neither number comes from a shared test, and Vapi's own docs say its displayed figure leaves out endpointing and transport. Test both with your carrier and script.

How do Retell AI, Vapi and ElevenLabs compare? ElevenLabs Agents sells plans with included minutes and charges $0.08 for each extra minute, plus the model; telephony comes from your carrier. At 10,000 minutes on its Pro plan that is about $812 before telephony, below Retell's $1,144 and at the bottom of Vapi's range.

Can I move an agent from Vapi to Retell? Not by import. Your prompts, tool definitions and webhooks move almost unchanged, but you rebuild the agent in Retell's format and re-point your numbers; numbers you own on Twilio, Telnyx or a SIP trunk can be imported into Retell.

If your support runs through a help desk rather than a phone line, you can start a Macha trial with $50 of free usage and no card.

Sources: Retell AI pricing · Retell vs Vapi (Retell) · Retell customers · Retell transfer call tool · Retell conversation flow · Vapi pricing · Vapi customers · Vapi warm transfer · Vapi knowledge bases · Vapi SIP trunking · Vapi chat quickstart · Vapi latency · Vapi Workflows migration · Vapi on AWS Marketplace · Trustpilot: Retell AI · Trustpilot: Vapi · Gartner Peer Insights: Retell AI · Capterra: Retell AI · Famulor: Retell AI vs Vapi · ElevenLabs Agents pricing