Vapi vs Retell AI: Notes From Running Both

Vapi vs Retell is the question most people ask once they decide to build a phone agent themselves instead of buying a finished product. Both platforms run the same chain on every call: speech to text, a language model, then a synthetic voice. Both bill by the minute and let you start with free credits. The difference is in what each one bundles, and in how much of the chain it lets you pick.

The short verdict: Retell puts more in the box. One per-minute price covers its voice engine, you get 20 concurrent calls free, and its docs list ready-made CRM and calendar connections plus a node-based conversation flow builder. Vapi charges a $0.05 per minute hosting fee, passes model costs through at cost, and lets you choose the provider for each layer, split a call across specialist assistants, and keep your whole setup as files in git. We run production agents on both at Voice Intego, and our public demo line runs on Vapi. Neither is the wrong choice. The right one depends on who is going to maintain the agent.

Vapi vs Retell at a glance

Everything in this table comes from each company’s own pricing page and documentation, read on October 30, 2026. Prices as of October 2026.

Sources: Vapi pricing, Retell AI pricing, Vapi docs introduction, Retell docs introduction.

What 450 minutes a month costs on each

Per-minute ranges are hard to compare in the abstract, so here is one realistic small business worked through. Say 150 inbound calls a month at an average of three minutes each. That’s 450 billed minutes.

Vapi, no package, using the rates on its own pricing calculator:

• Hosting: 450 x $0.05 = $22.50
• Transcriber (Deepgram, $0.0095 to $0.0099/min): $4.28 to $4.46
• Language model (OpenAI, $0.0077 to $0.0452/min): $3.47 to $20.34
• Voice (ElevenLabs, $0.0146 to $0.0238/min): $6.57 to $10.71
• Phone line: free on Vapi telephony, or 450 x $0.008 = $3.60 if the number sits on Twilio

Total: about $37 to $62 a month. One Vapi number is included, and four calls can run at once before you need to pay for more lines.

Retell, pay as you go, two setups from its component price list:

• Lean setup: voice engine $0.055 + Retell platform voice $0.015 + GPT 5.4 mini $0.024 + US telephony $0.015 = $0.109 a minute. 450 minutes comes to $49.05, plus $2 for a Retell number, so $51.05.
• Heavier setup: the same engine and telephony with ElevenLabs Flash ($0.040) and Claude 5 Sonnet ($0.064) = $0.174 a minute, or $78.30 for 450 minutes. With the number, $80.30.

So at this volume the Vapi vs Retell bill lands in the same range, and the model and voice you pick move the bill more than the platform does. Two billing details from Retell’s own FAQ are worth knowing before you estimate: silence and hold time are billed, because the transcriber keeps listening, and once a call transfers to a person the agent fee stops but the telephony fee carries on.

The bigger cost on either platform isn’t on the invoice. Someone has to write the prompt, wire up the booking tools, test on real phones and fix it when it drifts. If that person is you, count your hours. Our breakdown of what an AI receptionist costs covers the finished-product side of the same question.

How you build an agent on each

On Vapi the basic unit is an assistant: a single system prompt plus tools and structured outputs. When one prompt gets too crowded, Vapi’s answer is a Squad, where several assistants each handle part of the call and hand the caller between them with the context kept. Vapi’s docs also describe a CLI and a config-as-code setup where assistants, tools and tests live as files in git, so changes can be reviewed and tested before they reach production. You pick the transcriber, model and voice provider separately for each assistant, or start from one of its presets.

Retell recommends starting with a single prompt agent and only moving to a conversation flow when the call has to follow a fixed order of steps, such as a compliance script or patient intake, or when the prompt has grown too long for the agent to follow. A conversation flow splits the call into nodes, each with its own instructions. Retell’s docs list warm transfer, voicemail detection, a knowledge base built from your website or documents, and CRM sync to Salesforce, HubSpot, Dynamics 365, GoHighLevel and Zoho. It also sells US and Canada numbers directly, or you can bring your own over SIP.

Both platforms offer simulation testing before an agent takes a real call, and both have an API for people who would rather not touch the dashboard. If you want a refresher on what each layer of the chain does, how an AI receptionist works walks through a single call.

What running both has taught us

We build and run agents for clients on Vapi and Retell, and every agent has its own booking workflow behind it, kept separate per agent and per platform. That’s deliberate. A change to one agent’s booking logic can’t break another, and moving an agent between platforms doesn’t drag anything else with it. A few things we learned the hard way:

• Keep the greeting to one sentence on Vapi. A two-sentence first message sometimes came out with a short gap of dead air in the middle, which callers hear as the agent cutting out. Vapi hands the greeting to the voice engine a sentence at a time. One sentence fixed it.
• Vapi’s assistant update replaces the whole model block. We once sent only the field we wanted to change, and the system prompt and tools were gone. Now we send the full block every time and read it back afterwards.
• Rules get written twice. On both platforms the agent reads your system prompt and the descriptions on each tool’s parameters. If the two disagree, the agent follows whichever it read last. Fixing the prompt alone fixes nothing.
• In August 2026 a voice model change made phone audio worse on one of our agents. ElevenLabs eleven_turbo_v2_5 came out mangled on the phone leg where eleven_flash_v2_5 was clean, and it sounded fine in the browser test. Both platforms offer ElevenLabs voices, so the lesson travels: when audio gets worse, compare the settings against the last call that sounded right.
• Test on ordinary mobile and landline calls. An internet test line is a poor judge of how the agent sounds to a real caller.

None of this is a mark against either platform. It’s what maintaining a phone agent looks like week to week, and it’s the work you take on when you build one yourself. For how those booking and routing steps fit into a business’s calls, see our walkthrough of phone call automation.

Who should build it themselves

Building directly on Vapi or Retell makes sense if you have a developer, or you are one, and you’ll be around to maintain the agent after launch. At low volume the platform bill is small, as the 450-minute example shows. Agencies running agents for many clients have good reasons too: Retell’s docs describe running each client in its own workspace, and Vapi sells extra organizations at $20 a month each.

Pick Retell if you want more ready-made pieces, like CRM sync and a node-based flow builder, and a simpler single price per minute. Pick Vapi if you want to swap each provider independently, split complex calls across a Squad, or keep everything in version control.

Building it yourself is a poor fit if nobody on your team will own the prompt, the booking tools and the testing. Prompts and tool settings need changing whenever your hours, services or calendar change, and someone has to make those changes.

Where Voice Intego fits

Voice Intego is the done-for-you option. We design the call flows, write the prompt and the tool rules, connect your calendar, and test on real phone calls before launch. You don’t pick a model or a voice engine; you pay one flat monthly price and the agent goes live in 7 days.

Plans on our pricing page as of October 2026: Starter at $497 a month with 650 minutes ($0.50 a minute over), Professional at $997 a month with 1,500 minutes ($0.40 a minute over), and custom Enterprise plans for multi-location and high-volume businesses. Every plan connects to Google, Outlook or Cal.com calendars. Professional adds CRM connections, and Enterprise adds custom integrations and API access, including a custom connection to field service software like ServiceTitan when your account includes API access. You can cancel anytime. If you’re unsure whether it pays for itself, the missed call cost calculator works it out from your call volume and job value.

Who we are not for: if you have a developer with time to spare and your call volume is modest, building on Vapi or Retell directly will cost far less than our plans, and we’d tell you so on a call. The same goes if all you need is a message taken. Our work suits businesses that want calls answered and appointments booked without anyone in-house becoming the agent’s maintainer. More on what that includes is on our AI voice agent service page.

To hear a production Vapi agent before deciding anything, call the demo line on +1 870 569 0092. It is the live agent, running on the same kind of setup described above.

Frequently asked questions

Is Vapi cheaper than Retell?

At small volumes they come out close. In our 450-minute example Vapi landed at about $37 to $62 a month and Retell at about $51 to $80, and the gap depends mostly on the model and voice you choose. Vapi’s paid packages and HIPAA add-on change the picture at larger scale, so price your own setup on both pricing pages.

Which has lower latency, Vapi or Retell?

Vapi’s docs state sub-600ms response times. We haven’t run a controlled side-by-side test, and the model and voice you pick sit in the middle of every reply on both platforms, so test with the exact settings you plan to use.

Can someone without coding experience build on Retell or Vapi?

Both have dashboards for building without code: Retell says an agent can be built entirely from the dashboard, and Vapi’s quickstart builds a first agent in the dashboard. Getting it to book reliably into a real calendar and handle edge cases is where most of the time goes.

Which is better for an agency running many clients?

Both support it. Retell’s docs describe a workspace per client, and Vapi counts workspaces as organizations, with two included and more at $20 a month. We keep a separate booking workflow for every agent either way, so one client’s change never touches another’s.

If you’d rather have it built and maintained for you, book a demo with Voice Intego or call the demo line on +1 870 569 0092 to hear one working first. For a wider view of finished products, our AI receptionist companies comparison lists the main options with their published prices.