Retell AI vs Vapi for a Small Business Phone Agent: Real Costs, Limits and Which to Pick

Retell AI vs Vapi: what a small business should know

You have watched the demo videos, you have a business that loses calls, and now you are staring at two sign-up pages that look nearly identical. Retell AI and Vapi are the platforms behind most of the DIY phone agents documented on this site, and picking the wrong one costs you a weekend of rebuilding.

The short version: they are closer than the marketing suggests, the price difference is mostly an accounting trick, and the deciding factors are usually limits and compliance rather than per-minute cost. The rest of this guide gives you the real numbers and a short test that settles it for your business.

How each one prices a minute

Vapi advertises $0.05 a minute. That is the hosting fee only. The speech recognition, the language model and the voice are billed separately at provider cost, and so is telephony if you use Twilio. Vapi's own estimator for 1,000 minutes a month lands between $82 and $129, which works out to roughly $0.08 to $0.13 a minute, as of October 2026.

Retell advertises $0.07 to $0.31 a minute for AI voice agents, and that rate bundles the voice infrastructure, the model and the voice. Its example default configuration comes to $0.11 a minute (the model at $0.04, voice infrastructure at $0.055 and text to speech at $0.015). Telephony through Retell is $0.015 a minute, and a phone number is $2 a month.

So at 1,000 minutes, Retell's default example is about $110 plus $15 of telephony, call it $125, and Vapi's estimate is $82 to $129. Once you add a decent voice and a capable model, the two are within a few dollars of each other. The cheaper headline rate is not a reason to choose Vapi. The number that matters is the all-in cost per call for your own call mix, which you can only get by running both.

Where the limits bite

Concurrent calls

Vapi's base package allows 4 concurrent calls and 1 phone number, Core ($29 a month) allows 10 calls and 5 numbers, and extra lines cost $10 each a month. Retell allows 20 concurrent calls on pay-as-you-go and charges $8 a month per extra concurrent call beyond that. For a one-location shop, 4 simultaneous calls is plenty. For a dispatch line during a storm, or a campaign that places outbound calls in bulk, it is a real ceiling. Do the math on your busiest hour, not your average.

Knowledge bases and numbers

Retell includes 10 knowledge bases free and charges $8 a month for each extra. If your agent answers from a long FAQ or a price sheet, check how each platform wants that content loaded before you commit.

Compliance

Vapi lists a HIPAA add-on at $2,000 a month. That single line takes it out of the running for most small medical, dental and therapy practices unless you are already building at a scale where it makes sense. Retell's pricing page does not show a HIPAA fee in the part I checked, so ask its sales team for the current terms and get them in writing before you put patient calls through either platform. For a dental or medical front desk, this question comes before any other.

Which one is easier to start with

The honest answer is Retell for a non-technical builder, with a caveat. Comparison articles, including several written by vendors with their own interests, describe Retell as closer to no-code and Vapi as built for developers, with Retell also offering more native connectors for tools like Cal.com, Make, n8n and GoHighLevel. That fits what the listings on this site show.

The builds documented here split both ways. The HVAC receptionist uses Retell with n8n, Make and Twilio (builtwithagents.ai/strategy/ai-voice-receptionist-hvac-retell-ai-full-build). The salon receptionist sold for $3,000 uses Vapi with Make (builtwithagents.ai/strategy/salon-ai-receptionist-3000-vapi-make). The plumbing and pest control builds use Vapi with n8n or Make (builtwithagents.ai/strategy/pest-control-ai-voice-receptionist-vapi-n8n). So both work in production, and the choice mostly follows who built it and what they already knew.

If you can write a prompt and drag nodes in n8n or Make, you can use either. If you want the lowest friction, start on Retell. If you want to swap in your own speech or model providers piece by piece, Vapi's pass-through model gives you that control.

A 90-minute test that settles it

Step 1: Write one call script

Take your single most common call, such as a customer booking an appointment, and write the prompt once in plain text. Include your hours, your service area, what the agent may and may not say, and when to transfer to a person. You will paste the same text into both platforms.

Step 2: Build the same agent on both

Create an agent on Retell and one on Vapi with the same prompt, a similar voice, and the same model class. Use the free credits ($10 on Retell and $5 on Vapi as of October 2026) so this costs nothing. Connect both to the same test calendar through Cal.com or Google Calendar.

Step 3: Make 15 scripted calls on each

Call each agent yourself with the same list: a normal booking, an out-of-area caller, a caller who interrupts, someone with a background noise problem, a request to speak to a person, a hard-to-spell name, and a question the agent should refuse to answer. Write down what breaks.

Step 4: Compare the billed cost

After the test, look at the usage page on each platform and divide the total by the minutes you used. That is your real all-in cost per minute on your call mix, and it settles the price argument with data.

Step 5: Decide on limits and fit

Check your peak concurrent calls against each platform's limit, check whether you need HIPAA, and check which one's integrations match your calendar and CRM. If the two are tied, choose the one whose dashboard you found easier to debug, because you will spend most of your time there.

What it costs overall

On top of the platform fee, count the pieces. As of October 2026: Retell's phone numbers are $2 a month and telephony is $0.015 a minute. Vapi's telephony through its own SIP is free, Twilio inbound runs $0.008 a minute, and Telnyx $0.0055 a minute, charged by the provider. Add n8n (listed at 20 euros a month on its Starter plan billed annually) or Make for the booking logic, plus your Google Workspace or CRM. A small shop can usually run a single agent for under $150 a month in total at around 1,000 minutes, and much less at lower volumes. If you pay someone to build it, expect setup fees from a few hundred dollars to several thousand, according to the builders documented on this site.

When neither is the right answer

If you already run on Jobber, ServiceTitan or Housecall Pro, try the built-in AI call answering before you build anything. The vendor handles booking inside your software, and you skip the integration work. If you have under 100 calls a month, a human answering service or a simple missed-call text back is cheaper and less to maintain. And if you cannot commit to reading transcripts for the first two weeks, do not put an agent on your phone line at all. An unmonitored agent will eventually say something you would not.

Frequently Asked Questions