Vapi vs Retell AI: Which Voice Agent Platform Fits You?
November 18, 2026

Vapi vs Retell AI is mostly a question of how much of the voice stack you want to choose yourself. Vapi is an API-first platform: it charges a per-minute hosting fee, passes model costs through, and lets you plug in your own provider keys. Retell AI sells one itemised per-minute rate, ships a visual conversation flow builder, and includes more capacity before you pay for anything extra.
The short version:
- Pick Vapi if you're a developer who wants to choose the transcriber, the model and the voice separately, manage agents as code, or bill providers directly with your own keys.
- Pick Retell AI if you want a visual, node-by-node builder for calls that have to follow a script, and one invoice that covers the whole call.
- Test both before you decide on speed. Latency depends on the model, voice, region and phone carrier you pick, so no number on a marketing page will match your calls.
One warning about this search. Look up "Retell vs Vapi" and many of the results are Retell's own comparison pages, or posts from other voice platforms that compete with both. So everything below was checked on each vendor's own docs and pricing page on November 18, 2026, not on anyone's comparison page.
Vapi vs Retell AI: quick comparison
| Vapi | Retell AI | |
|---|---|---|
| Building blocks | Assistants and Squads (multi-assistant handoffs) | Single prompt agents and conversation flow agents |
| Visual flow builder | No. Workflows was retired in August 2026 | Yes, conversation flow |
| Pricing model | Per-minute hosting fee plus model, voice, transcriber and telephony costs at provider cost | One per-minute rate built from voice infrastructure, LLM, text-to-speech and telephony |
| Own provider keys | Yes, for transcriber, model and voice providers | No self-serve option found in its docs. Custom LLM over your own WebSocket server |
| Concurrent calls included | 4 on usage-only | 20 on pay-as-you-go |
| Phone numbers | One included US number (inbound only); import from Twilio, Telnyx, DIDWW or a SIP trunk | Numbers from Retell, or import through SIP trunking |
| Compliance, as stated | HIPAA, SOC 2 Type II, GDPR; HIPAA is a paid add-on | HIPAA, SOC 2 Type 1 and 2, GDPR; BAA self-signed at no extra fee |
| EU hosting | EU region, on the top support package only | Says it does not currently operate services in the EU |
| On-premise | No, per its FAQ | Not mentioned in the docs we checked |
How you build an agent
This is where the two products have drifted furthest apart in 2026, and where older comparison posts are most often wrong.
Vapi: Assistants and Squads, no more Workflows
A Vapi assistant is one prompt, one model, one voice and a set of tools. For anything bigger, you use Squads: several focused assistants that hand the call to each other, for example one that qualifies a lead and another that books the appointment. Vapi's Squads docs pitch this as a fix for long, do-everything prompts.
What Vapi no longer has is a visual node-and-edge builder. Its Workflows migration guide says Workflows would be retired on August 18, 2026, that existing workflows stop running from August 19, and that Squads is the recommended replacement. Vapi's stated reason is blunt: models weren't reliably keeping track of the current node and its possible next steps, and the Squads pattern gave customers better results.
If a blog post or video shows you a Vapi flowchart canvas, it's out of date. Today you describe behaviour in prompts and handoff rules, in the dashboard, through the API, or as config files in Git.
Retell AI: single prompt first, conversation flow when you need control
Retell gives you two agent types, and its docs are refreshingly honest about choosing between them. The single prompt vs. conversation flow guide says to start with a single prompt agent and move to conversation flow only when the simple version stops meeting your needs.
Conversation flow is the visual option. You split the call into nodes (dialogue, function calls, transfers, logic splits, code), then define the conditions for moving between them. Retell says this helps when you need a more deterministic order of steps, which matters for compliance scripts, but it also says flow "is not automatically more reliable" and that extra nodes can make a simple call harder.
One change to know about: Retell is retiring its older multi-prompt agents. Its deprecation notice says new multi-prompt creation ends on December 15, 2026 and editing ends on February 1, 2027, with conversation flow as the replacement.
What that means for you: the two vendors made opposite bets. Vapi dropped its graph builder in favour of prompt-driven handoffs. Retell is consolidating on its graph builder. If your team thinks in flowcharts, or legal needs to sign off on a call script step by step, Retell matches how you work. If you'd rather keep logic in prompts and code, Vapi does.
Pricing: a hosting fee plus pass-through vs one itemised rate
Both platforms bill by the minute, and both pricing pages break the minute into parts. The difference is who sends the bill for each part.
Vapi charges its own hosting fee, listed at $0.05 per minute on the Vapi pricing page, and says models are passed through at provider cost. On top of that you pay for the transcriber, the language model, the voice and the phone carrier. Add your own provider keys and those providers bill you directly instead.
Vapi then sells optional Success Packages (Core, Pro and Premier). They don't change usage billing. They add support response commitments, more concurrent call lines, longer data retention and, at the top tier, extras like SSO.
Retell AI publishes a single pay-as-you-go range of $0.07 to $0.31 per minute for voice agents on the Retell pricing page. That range is the sum of Retell's voice infrastructure fee, the LLM you pick, the text-to-speech voice, and telephony. Retell bills all of it, and its calculator shows the per-minute total as you change options.
| What you pay for | Vapi | Retell AI |
|---|---|---|
| Platform | $0.05/min hosting fee | $0.055/min voice infrastructure |
| LLM, voice, transcription | Provider cost, or direct with your own keys | Per-minute price per model and voice, set by Retell |
| Starting credit | $5 | $10 |
| Concurrent calls included | 4, extra lines sold monthly | 20, extra lines sold monthly |
| Support and uptime commitments | Sold as Success Packages | Enterprise plan |
| HIPAA | $2,000/month add-on | BAA at no additional fee |
Those figures come from each pricing page on the date above and they change often, so open both pages before you budget.
A fair way to compare: price one realistic call on both calculators with the same model and voice. Vapi's headline fee looks lower, but it's only the platform's slice. Retell's range looks higher, but it already includes the model and the voice. Depending on your choices, either one can come out ahead.
The line items that surprise people are elsewhere. On Vapi, check concurrency (4 lines on usage-only) and the HIPAA add-on. On Retell, check the per-minute add-ons such as knowledge base, denoising and PII removal, which stack on top of the base rate.
Telephony, models and keys
Phone numbers
Vapi gives you one US phone number at no charge, and its phone calling docs are clear that these numbers only receive calls. For outbound or international calling you import a number from Twilio, Telnyx or DIDWW, or connect a SIP trunk.
Retell sells phone numbers itself for a monthly fee and lists a per-minute telephony charge for calls on its Twilio and Telnyx-backed numbers. You can also bring your own carrier: Retell's custom telephony guide covers elastic SIP trunking and dialling to a SIP URI, and the pricing page says there is no charge for SIP trunking or custom telephony.
Both will connect to an existing contact-centre phone system over SIP. Retell is quicker if you have no carrier at all and need to call out on day one.
Bring your own models and keys
This is Vapi's clearest advantage. Its provider keys docs say you can add your own API keys for transcription, model and voice providers, and that once a key is validated you're charged by that provider instead of through Vapi. That matters if you have volume pricing with a model vendor or a custom voice. Vapi also supports custom LLMs through an OpenAI-compatible endpoint.
Retell's route is different. You choose from the models and voices on its price list, or you run a custom LLM: Retell opens a WebSocket to your server for each call, sends the live transcript, and speaks whatever you stream back. Retell's custom LLM overview is frank about the cost of that choice. You take on latency and uptime, and you lose the LLM playground, simulation testing and agent transfer for those agents.
We couldn't find a Retell doc describing a self-serve way to plug in your own provider API keys the way Vapi does, so we're not claiming it exists. If that's a requirement, ask Retell what an Enterprise agreement covers.
Compliance and data residency
These are the vendors' own statements, not our audit. Ask for the reports before you sign.
- Vapi says in its FAQ that it is HIPAA compliant, SOC 2 Type II-attested and GDPR compliant. HIPAA handling, including a BAA, is sold as a monthly add-on. Vapi has an EU region, but its docs say that region is available only with the Premier package. The same FAQ says Vapi does not support on-premise deployments.
- Retell AI says on its security and compliance page that it is HIPAA compliant, GDPR compliant and SOC 2 Type 1 and Type 2 certified. Its BAA and DPA can be self-signed online, with no additional fee. The same page notes that Retell does not currently operate services within the European Union, and that it doesn't train models on data from enterprise customers with a master services agreement.
For a small healthcare team, the cost of HIPAA is a real difference: a fixed monthly add-on on one side, a self-signed agreement on the other. For an EU company that needs calls processed in-region, read both vendors' pages closely, because neither offers that on a self-serve plan.
Retell vs Vapi: which one should you pick?
| Your situation | Better fit |
|---|---|
| You want to pick each provider, or use your own keys and negotiated rates | Vapi |
| You manage agents as code and review changes in Git | Vapi |
| You need a visual flow that a non-engineer can read and approve | Retell AI |
| You want one bill and more concurrent calls without add-ons | Retell AI |
| You handle patient data on a small budget | Retell AI, after reading its BAA |
| You need on-premise or full control of the media pipeline | Neither is built for it. Vapi says no on-premise; look at open frameworks |
If neither fits, the Vapi alternatives and Retell AI alternatives roundups cover the wider field, and our LiveKit alternatives post covers the lower-level route, with WebRTC infrastructure and open voice agent frameworks.
Not sure a business call agent is what you need at all? Our guide to AI voice assistants sorts phone assistants, call screening and business call agents by the job each one does.
Where Taku fits
Vapi and Retell AI are both platforms for building a voice agent yourself. If you'd rather run a tool someone already built than set one up yourself, that's where Taku comes in. Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.
FAQ
Is Vapi cheaper than Retell AI?
It depends on what you plug in. Vapi lists a lower platform fee, but you pay for the model, voice, transcriber and carrier on top. Retell's published range already includes those parts. Price the same call on both calculators before you compare.
Does Vapi still have a visual workflow builder?
No. Vapi's docs say Workflows would be retired on August 18, 2026 and that existing workflows stop running after that date. Vapi now recommends Assistants for simple agents and Squads for multi-step ones.
Which is faster, Retell or Vapi?
Neither vendor's number will match your setup. Response time changes with the LLM, the voice, the speech recognition provider, your region and the phone carrier. Build the same short agent on both, call it from a real phone, and measure.
Can I use my own LLM with Retell AI?
Yes, through its custom LLM option, where Retell connects to your WebSocket server on each call. Retell's docs recommend its built-in agent types for most teams, because custom LLM agents lose some testing and transfer features.
Are Vapi and Retell AI HIPAA compliant?
Both say so in their docs. Vapi sells HIPAA handling with a BAA as a paid monthly add-on. Retell says a signed BAA is required before you send patient data and that signing it carries no additional fee. Get each vendor's compliance reports and have your own counsel review them.
Can I move from one to the other later?
Yes, but expect a rebuild, not an import. Prompts and tool definitions carry over fairly well. Vapi Squads and Retell conversation flows are structured differently, so the call logic has to be redesigned.