If you build voice agents for clients, the Vapi-vs-Retell question isn't really about which product has the better demo. It's about which one you can bill against, scale across accounts, and keep live without babysitting. We've shipped client agents on both, so here's the short answer up front: pick Retell when you want a predictable per-minute rate you can resell and unlimited concurrency for campaigns; pick Vapi when you have developers and genuinely need to own every layer of the stack. For most marketing and automation agencies, that lands on Retell — but not for the reason the pricing pages suggest.
The rest of this is the agency-side reasoning: what the headline numbers hide, where each platform actually bites, and the checklist we run before we standardize on one for a client.
The $0.05 Trap: Why the Cheapest Headline Rate Isn't the Cheapest Bill
Vapi advertises a $0.05/min platform fee — the lowest base rate among the major platforms. That number is true and almost useless on its own.
Vapi's model unbundles everything. On top of that $0.05 platform fee you're separately paying for speech-to-text (~$0.01/min), the LLM (~$0.02–$0.20/min depending on model), text-to-speech (~$0.04/min), and telephony (~$0.01/min). Add it up and a real deployment lands somewhere around $0.13 to $0.31+ per minute — and you're reconciling four or five vendor bills to get there.
Retell charges $0.07+/min with no separate platform fee, and bundles the components into one consolidated number. A typical Retell agent running GPT-4o, Deepgram for transcription, and ElevenLabs for voice comes out around $0.11–$0.15/min on a single invoice.
So which is cheaper? It depends entirely on whether you'll do the optimization work:
| Factor | Vapi | Retell | |---|---|---| | Headline platform fee | $0.05/min | $0.07+/min (no platform fee) | | Realistic all-in cost | ~$0.13–$0.31/min | ~$0.11–$0.15/min | | Billing | Component-level, multiple vendors | Consolidated, one invoice | | Free tier | ~1,000 min/month | Pay-as-you-go | | Best when | You optimize each component | You want one predictable number |
For an agency reselling minutes, the consolidated number wins more often than the low headline. You can't put a clean margin on a rate that swings $0.13 to $0.31 depending on which LLM the agent happened to route to that call. We wrote about the underlying math in more depth in what an AI voice agent really costs per minute — the pattern there is the same: the number on the pricing page is the start of the calculation, not the end.
Concurrency Is the Feature Nobody Compares (And It Decides the Winner)
Here's the difference that actually changes what you can sell. Retell supports unlimited concurrent call capacity. Vapi's concurrency is more often gated behind higher-tier plans.
This sounds like a footnote until you run an outbound campaign. Say a client wants 500 leads called in a two-hour window, or an inbound line that can't drop calls during a promo spike. On Retell, you dial up the volume and the platform absorbs it. On Vapi, you can hit a concurrency ceiling that forces a plan upgrade mid-campaign — exactly when you least want a surprise.
For a single-app developer, concurrency limits are manageable. For an agency running one platform account across many client agents, they're a scaling tax. The whole agency model is: build once, resell minutes, handle everyone's spikes from the same account. Unlimited concurrency is what makes that model hold together. This is the same reason we lean on it for lead qualification agents that field inbound bursts — the calls don't arrive politely spaced out.
Build Effort: Own-Every-Layer vs Ship-This-Week
The second real fork is how much engineering you want to carry.
Vapi is built for teams that want control of every layer — custom function calling, their own LLM routing, granular telephony behavior. It rewards a staff of developers who'll wire that up and maintain it. If you have that team and a reason to own the stack (a proprietary product, unusual telephony needs, per-call model routing you want to tune), Vapi gives you the room.
Retell leans the other way: ease of use, templates, and integrations that get a working agent live without a dedicated engineer. You can stand up an inbound qualifier or an appointment booker in days, not sprints. For a lean agency — where the goal is a live, revenue-generating client agent this week — that speed is the whole game.
Neither is "no-code," to be clear. You'll still design the conversation, wire the tools, and QA real calls. But the floor of effort to get to a working client agent is meaningfully lower on Retell. If you want the honest version of why voice deployments fail regardless of platform, most of it isn't the platform at all — it's prompt and workflow design, which we broke down in why AI SDRs fail.
The White-Label Question Agencies Always Ask
Can you brand this as your own service? Yes — with the right architecture.
Neither Vapi nor Retell hands you a fully branded end-client dashboard out of the box. The move that works is to keep the platform account internal and expose a thin client-facing layer: transcripts, call outcomes, booking data, and a simple results view, pushed into the client's CRM or a lightweight portal. The client sees outcomes, not the platform. That keeps the underlying vendor invisible, keeps the data yours, and — critically — keeps the margin yours.
This is also why the platform choice matters less to the client than it does to you. They're buying "qualified leads booked on the calendar," not "a Retell agent." Your job is to make the plumbing disappear. If you're standing up that whole delivery model from scratch, our agency guide to running voice agents as a service walks through the wrapper layer.
Where Both Platforms Still Bite (Regardless of Which You Pick)
Choosing the platform is the easy 20%. Here's what actually determines whether a client agent works, and it's mostly platform-agnostic:
- Latency compounds. Every layer — transcription, LLM, voice — adds delay, and the caller feels the total. A cheaper LLM that thinks half a second longer will tank perceived quality faster than a slightly higher per-minute rate ever will. Test end-to-end latency on real phone calls, not in the platform's web widget.
- Telephony is where calls die. Number provisioning, carrier filtering, and spam labeling will break more deployments than the AI ever does. Both platforms lean on Twilio-style telephony under the hood; budget time for warming numbers and handling "scam likely" flags on outbound.
- The prompt and call flow are 80% of quality. We've watched the same client agent go from unusable to closing appointments with zero platform change — just a rewritten flow, better interruption handling, and tighter fallback logic. Don't blame the platform for a design problem.
- Data handoff is the deliverable. A call that isn't logged into the CRM with a clear outcome is a call the client can't see, and therefore can't value. Wire the outcome pipeline before you scale the call volume.
None of these are Vapi-vs-Retell decisions. They're the reason two agencies on the same platform can get wildly different results — and why "which platform" is the wrong question to obsess over past the first pilot.
So Which One Do We Reach For?
For most marketing and automation agencies, Retell is the safer default: a predictable per-minute number you can mark up and resell, unlimited concurrency that survives campaigns, and faster time-to-live for a client agent. That's not a knock on Vapi — it's a match to how agencies actually make money on voice.
We reach for Vapi when a client genuinely needs to own the stack: a proprietary product, unusual per-call model routing, or a developer team that will optimize each component and actually realize that lower headline cost. In that world, Vapi's control is worth the extra plumbing.
The one bad reason to pick Vapi is the $0.05 headline. That number almost never survives a real deployment, and if it drove your decision, you'll feel it on the first client invoice.
You can validate the official specs and pricing directly on the Vapi docs and Retell docs — but read them as component sheets, not as your final cost. Your final cost is set by how you bill, how much you scale, and how much engineering you're willing to carry.
Want Us to Pick and Build It for You?
We deploy voice agents for clients on both platforms and choose based on your billing model, call volume, and how you want to package it — not on whichever vendor has the flashier landing page. If you're weighing Vapi vs Retell for an inbound qualifier, an outbound booker, or a front-desk agent, we'll size it honestly and tell you which one fits.
Get a Free Automation Audit and we'll map the right voice stack — platform, cost per minute, and the wrapper that makes it look like yours — to your actual client model.