Vapi vs Retell AI: Which AI Voice Agent Platform Wins in 2026?

If you are building an AI phone agent in 2026, two platforms end up on almost every shortlist: Vapi and Retell AI. Both give you a production voice agent that can answer calls, book appointments, qualify leads, and hand off to a human. Neither is a thin wrapper around a single model.

But they are built on opposite philosophies. Vapi is an orchestration layer that hands you every component and lets you assemble the stack yourself. Retell AI is an all-in platform that bundles the stack and prices it as one number.

That difference shows up everywhere — in your monthly bill, your compliance timeline, and how many engineers you need to keep the thing running. This comparison breaks down the real numbers from both vendors' published pricing, current as of September 2026.

---

Quick Comparison: Vapi vs Retell AI

Factor	Vapi	Retell AI
Pricing model	Platform fee + unbundled provider costs	All-in per-minute, no platform fee
Base rate	$0.05/min hosting	$0.055/min voice infrastructure
All-in range	~$0.10–$0.14/min (derived from component rates)	$0.07–$0.31/min (published)
Free credits	$5	$10
Free concurrency	4 calls	20 calls
Extra concurrency	$10/line/mo	$8/concurrency/mo
Entry paid plan	$29/mo (Core Success Package)	None required — pay as you go
HIPAA	$2,000/mo add-on	Self-serve BAA, no extra fee
Zero data retention	Paid add-on (~$1,000/mo, dashboard-quoted); excludes HIPAA mode	Configurable retention included
Certifications	SOC 2 Type II, SOC 3, GDPR, PCI DSS	SOC 2 Type I and II, HIPAA, GDPR
Bring your own keys	✅ Full BYOK across STT/LLM/TTS	Partial (custom telephony free)
Latency	~600ms claimed (3.08s measured by Cekura)	~600ms claimed (2.21s measured by Cekura)
Best for	Engineering teams tuning every component	Teams shipping fast in regulated markets

---

What Is Vapi?

Vapi is a developer-first voice AI orchestration platform. You pick a transcriber, a language model, and a voice provider, then Vapi wires them together with telephony and handles the hard real-time parts: streaming audio, turn detection, interruption handling, and tool calls.

The core building block is an assistant — a system prompt plus a set of tools that can hit your APIs and databases. For anything more complex, Vapi offers Squads, a multi-assistant orchestration pattern where specialised agents transfer calls between each other. A booking agent hands off to a billing agent without dropping the caller.

Vapi integrates dozens of providers including OpenAI, Anthropic, Google, Deepgram, Gladia, and ElevenLabs. It targets sub-600ms response times with natural turn-taking, which is roughly the threshold where a phone conversation stops feeling like a phone tree.

There is a CLI, web SDKs for embedding voice calls directly in your app, and inbound plus outbound phone calling.

---

What Is Retell AI?

Retell AI is an all-in voice agent platform aimed at teams that want a working phone agent without assembling the pipeline themselves. Its headline positioning is "no platform fees, no feature gating" — every plan gets the full product.

Retell gives you two agent types. A Single Prompt Agent is the fast path for simple use cases. A Conversation Flow Agent splits the task into discrete nodes, which gives you tighter execution control and less context per node — useful when an agent has to follow a script reliably rather than improvise.

Beyond the builder, Retell bundles things Vapi expects you to source or build:

• Knowledge base — crawl a website or upload documents the agent retrieves from
• Simulation testing — run graded test cases to catch regressions before launch
• Playground — chat, call, or debug an agent interactively
• Built-in analytics — call success and sentiment scoring, plus custom dashboards
• Prebuilt integrations — CRM, helpdesk, calendar, and file storage, connected once and available to every agent

There are official Node.js and Python client libraries, plus an MCP server so you can drive Retell from tools like Cursor or Claude Desktop.

---

Pricing: The Number That Actually Matters

Both vendors advertise a low headline rate. Neither headline rate is what you pay. Here is how the stacks actually build up from published component pricing.

Vapi's cost stack

Vapi charges $0.05/min for hosting, then passes through each provider separately:

Component	Published rate
Vapi hosting	$0.05/min
Transcriber (Deepgram)	$0.0095–$0.0099/min
Model (OpenAI)	$0.0077–$0.0452/min
Voice (ElevenLabs)	$0.0146–$0.0238/min
Twilio outbound	$0.014/min

A cheap Vapi stack lands around $0.096/min. A premium model stack lands around $0.143/min. Transport via Vapi's own telephony, SIP, WebSockets, or Daily WebRTC is free, so routing calls through Vapi instead of Twilio shaves roughly $0.014/min off outbound.

Retell AI's cost stack

Retell splits into infrastructure, voice, telephony, and model:

Component	Published rate
Retell voice infrastructure	$0.055/min
Text-to-speech (platform voices)	$0.015/min
Text-to-speech (ElevenLabs)	$0.040/min
Telephony	$0.015/min (custom telephony free)
GPT 5 nano	$0.0016/min
GPT 5.6 Terra / Claude 5 Sonnet	$0.064/min
GPT 5.5	$0.16/min

A budget Retell stack — platform voice, Retell telephony, GPT 5 nano — comes to about $0.087/min. A premium stack with Claude 5 Sonnet and ElevenLabs voices reaches about $0.174/min.

The verdict on price

At the low end the two platforms are close enough that price should not decide it. Vapi's cheap stack ($0.096) and Retell's cheap stack ($0.087) are within a cent of each other.

The gap opens on model choice. Retell charges $0.064/min for a frontier model where Vapi's OpenAI pass-through tops out at $0.0452/min. If your agent genuinely needs a premium LLM at high volume, Vapi's pass-through pricing plus BYOK is the cheaper path.

Vapi's real cost lever is BYOK. When you supply your own provider API keys, Vapi does not bill you for that provider at all — the provider bills you directly. If you already have enterprise volume pricing with OpenAI or ElevenLabs, that discount flows straight through, and Vapi's effective cost collapses toward the $0.05/min hosting fee.

---

Concurrency: An Underrated Deal-Breaker

This is where the two platforms diverge most sharply, and it catches teams off guard.

Vapi's free tier allows 4 concurrent calls. The $29/mo Core Success Package raises that to 10. Pro — which costs a $999/mo minimum, billed as 10% of your hosting fee — gets you 30. Additional lines are $10/line/month.

Retell gives you 20 free concurrent calls on pay-as-you-go, with additional concurrency at $8/concurrency/month.

Run the comparison at 20 concurrent lines. On Retell, that is included. On Vapi, you are on the Core plan ($29) plus 10 extra lines ($100) — $129/month before a single minute of audio.

If you are running outbound campaigns or have spiky inbound traffic, Retell's concurrency allowance is worth more than any per-minute difference.

---

Compliance: Retell's Clearest Win

Both platforms are SOC 2 Type II certified. Vapi has completed annual SOC 2 Type II and SOC 3 examinations covering July 1 2025 – July 31 2026, and documents GDPR and PCI DSS v4.0.1 compliance. Retell's documentation lists SOC 2 Type 1 and Type 2, HIPAA, and GDPR.

One caveat worth knowing if ISO 27001 is on your procurement checklist: Retell does not hold it. Its marketing pages have displayed an ISO 27001 badge, but its own compliance documentation lists only SOC 2, HIPAA, and GDPR. Ask for the certificate before assuming coverage. Retell also notes it does not currently operate services inside the EU, which matters if you need in-region data residency rather than GDPR compliance via a DPA.

The difference is how you get to a signed agreement.

Retell's BAA and DPA (including EU Standard Contractual Clauses) are self-signable, with no additional fee for signing. A signed BAA is required before you transmit PHI. Retell also offers automatic PII redaction (+$0.01/min), configurable data retention, and on-premises deployment. Note that SSO and module-level role-based access control sit on the Enterprise plan rather than pay-as-you-go — budget for a sales conversation if you need either.

Vapi gates HIPAA behind a $2,000/month add-on — $24,000/year before you place a single call. Zero data retention is a separate paid add-on, widely reported at around $1,000/month, though that figure is quoted in-dashboard rather than published on the pricing page.

Importantly, you cannot stack them. Vapi's documentation states that HIPAA mode and Zero Data Retention are mutually exclusive — you must disable one to enable the other. So $2,000/month is the realistic compliance ceiling, not $3,000.

For a healthcare clinic running 5,000 minutes a month, the compliance overhead alone decides this. Retell at ~$0.10/min with PII removal is roughly $500/month. Vapi at a comparable rate plus the HIPAA add-on is roughly $2,500/month. The audio costs the same; the paperwork costs five times more.

If you handle PHI, financial records, or EU personal data and you do not have a dedicated voice AI engineering function, Retell's bundled compliance is the pragmatic choice.

---

Quality and Latency

Both vendors market roughly 600ms turn-taking latency, which is about the threshold where conversation feels natural rather than transactional. Treat those as marketing figures, not measurements.

Independent testing tells a different story. Cekura, which runs third-party voice AI benchmarks across platforms including Vapi, Retell, LiveKit, Pipecat, and ElevenLabs, measured mean response times of 2.21 seconds for Retell and 3.08 seconds for Vapi — four to five times the advertised numbers, and a real gap in Retell's favour. Vendor latency claims typically measure one hop in the pipeline rather than the full round trip a caller actually experiences, so expect real-world performance closer to Cekura's figures than the marketing page.

On conversational quality, Cekura's published results show Retell leading its Agent Workflow benchmark at 75.61% pass for repeatability across 8 platforms and 82 scenarios. Retell also scored 92.2 on voice quality — strong, though second place there, behind Bland at 92.8.

Retell separately cites a Cekura result of 95.7% accuracy across 414 calls, the highest of six platforms tested. Two caveats: that figure is specific to Cekura's Medicare workflow test rather than general workflow accuracy, and it is a vendor-selected citation. The underlying independent benchmark does genuinely place Retell at the top of the workflow category, so the direction holds even if the headline number is narrower than it sounds.

The honest read: Retell has a measurable edge on out-of-the-box workflow reliability. Vapi's ceiling depends on how well you tune your own stack, which is both its weakness and its point.

---

Who Should Use Vapi?

Choose Vapi if you:

• Have an opinion about every component — you want to pick the STT, the model, and the voice, and swap them independently
• Already have provider volume discounts — BYOK turns your OpenAI or ElevenLabs enterprise rate into a direct cost advantage
• Run high volume with a tuned cheap stack — at scale, unbundled pass-through pricing beats bundled rates
• Need multi-agent orchestration — Squads handles specialised agent handoffs natively
• Have engineering capacity to own the pipeline — Vapi gives you levers, not defaults

Realistic fit: A product team embedding voice in their own app, running 100,000 minutes a month on a tuned stack with their own OpenAI contract. At $0.05/min hosting plus BYOK provider costs, Vapi's unit economics are hard to match.

---

Who Should Use Retell AI?

Choose Retell if you:

• Operate in a regulated market — self-serve BAA at no extra cost versus $2,000/month
• Need predictable costs — one all-in number you can put in a budget
• Want testing and analytics included — simulation tests, sentiment scoring, and dashboards ship with the product
• Run concurrent campaigns — 20 free concurrent calls versus 4
• Are shipping without a dedicated voice AI engineer — the conversation flow builder and playground do a lot of the work

Realistic fit: A multi-location dental group replacing an after-hours answering service. 8,000 minutes a month, PHI in scope, no in-house engineering team. Retell with PII removal runs roughly $800/month and can be live in days.

---

Vapi vs Retell AI: Verdict

Retell AI wins for most teams in 2026. Bundled compliance, 20 free concurrent calls, included testing and analytics, and a single forecastable per-minute rate remove most of the operational drag from shipping a voice agent. The benchmark data on workflow reliability supports it too.

Vapi wins when you are optimising unit economics at scale. The $0.05/min hosting fee is among the lowest platform rates in the category, and BYOK means your provider discounts pass straight through. If you have the engineering function to tune and maintain the stack, Vapi's ceiling is higher and its floor is cheaper.

The decision reduces to two questions:

1. Do you handle regulated data? If yes, Retell's self-serve BAA versus Vapi's $2,000/month add-on likely settles it.
2. Do you have engineering capacity to own a voice pipeline? If no, Retell's defaults are better than the stack you will assemble under time pressure.

Still mapping the category? See our roundup of the best AI agent platforms for the broader landscape, or our AI customer support tools guide if phone support is the specific problem you are solving.

Pricing verified September 2026 against both vendors' published pricing pages, except Vapi's zero-data-retention fee, which is dashboard-quoted and sourced from third-party reports. Benchmark figures are from Cekura's independent testing. Voice AI pricing changes frequently — confirm current rates before committing.

---

Frequently Asked Questions

Is Vapi or Retell AI cheaper? At the low end they are nearly identical — roughly $0.087/min for a budget Retell stack versus $0.096/min for a budget Vapi stack. Vapi becomes cheaper at high volume if you bring your own provider keys. Retell is cheaper once you factor in concurrency and compliance, which Vapi charges extra for.

Does Vapi have a free tier? Yes. Vapi offers $5 in free credits with 4 concurrent calls, 1 phone number, 14-day data retention, and community Discord support.

Is Retell AI HIPAA compliant? Yes. Retell is HIPAA compliant and its BAA is self-signable at no additional cost. A signed BAA is required before transmitting PHI. Vapi charges $2,000/month for HIPAA compliance as an add-on.

Can I use my own OpenAI API key with Vapi? Yes. Vapi supports bring-your-own-keys across transcription, model, voice, and cloud storage providers. When you supply your own key, Vapi does not bill you for that provider — the provider bills you directly.

Which platform has lower latency? Both advertise roughly 600ms turn-taking, but independent Cekura benchmarks measured 2.21s mean response for Retell versus 3.08s for Vapi — so Retell is measurably faster in practice, and both are slower than their marketing suggests.

How many concurrent calls do I get? Retell includes 20 concurrent calls on pay-as-you-go, with extra concurrency at $8/month each. Vapi allows 4 on the free tier, 10 on the $29/mo Core plan, and 30 on Pro, with extra lines at $10/month each.

Do I need a Vapi paid plan? Not to start — hosting is usage-based and prepaid. But the free tier's 4-call concurrency limit and 14-day retention push most production deployments onto at least the $29/mo Core package.