Retell AI vs. Vapi: component pricing and provider control

Retell AI and Vapi are developer-oriented voice-agent platforms with usage-based entry plans. Both let teams choose parts of the speech and model stack. The pricing mechanics are different enough that a headline rate does not settle the decision: Retell publishes prices for each major component, while Vapi publishes a hosting fee and passes speech, model, and voice costs through at provider cost or lets customers bring their own keys. Retell pricing, accessed 2026-08-19; Vapi pricing, accessed 2026-08-18.

That distinction drives a Retell vs. Vapi pricing comparison. Retell's configured total can be assembled from its public rate card. Vapi's hosting line can be calculated, but its complete call price depends on provider and telephony charges not priced on Vapi's public page. Provider control, testing, concurrency, retention, and compliance fees can outweigh a small difference in the platform line.

A third shape is an all-in engine rate without separately priced speech, model, and voice components. ThunderPhone, the publisher of this page, is one and appears after the two named vendors, disclosed under the same rules. Every third-party claim below carries its first-party link and the date it was accessed.

Retell AI: published component prices

Retell is pay as you go with no mandatory subscription, and new accounts start with $10 in free credits. Every voice minute pays a $0.055 infrastructure charge. The language-model line ranges from $0.003 per minute for GPT 5 nano to $0.16 for GPT 5.5 among standard models, while listed fast variants reach $0.32 per minute. Text-to-speech is $0.015 per minute for Retell platform voices, MiniMax, Fish, Cartesia, and OpenAI, or $0.040 for ElevenLabs. U.S. telephony through Twilio is $0.015 per minute, while Retell charges nothing for SIP trunking with custom telephony. A Retell number is $2 per month; a verified number is $10. Retell pricing, accessed 2026-08-19.

Retell advertises a $0.07–$0.31 per-minute range, with the headline excluding the LLM and telephony. Its component table is the more useful budgeting source because configurations can go above that range. A standard GPT 5.5 and ElevenLabs setup with U.S. telephony computes to $0.27 per minute before a number, while GPT 5.5 fast in the same stack computes to $0.43. Retell pricing, accessed 2026-08-19.

Optional features add their own usage lines. Knowledge base, advanced denoising, and safety guardrails each add $0.005 per minute; PII removal adds $0.01; AI Quality Assurance costs $0.10 per minute after the first 100 free minutes; batch calling adds $0.005 per dial. Twenty concurrent calls are included, then added concurrency costs $8 per line per month. Enterprise offers custom pricing, volume discounts, and uncapped concurrency. Retell pricing, accessed 2026-08-19.

Retell says language support depends on the voice and speech-recognition providers, with per-provider coverage running roughly 30–80 languages and the dashboard serving as the source of truth. A multilingual combination must be supported by one voice and one recognition provider. Telephony supports existing numbers and custom SIP, with native Twilio, Vonage, and Telnyx integrations. Retell language docs and Retell homepage, accessed 2026-08-18.

Its compliance docs state SOC 2 Type 1 and Type 2, HIPAA with a signed BAA required before transmitting PHI, and GDPR through a DPA and standard contractual clauses, while noting that Retell does not currently operate services within the European Union. Retell also documents LLM simulation testing, test cases, batch runs, and mocks. Retell compliance docs and simulation-testing docs, accessed 2026-08-18.

Vapi: hosting plus provider pass-through

Vapi's Build plan charges $0.05 per minute for call hosting and includes “60+ minutes,” though the pricing page does not state whether that allowance is one-time or monthly. Speech-to-text, the language model, and text-to-speech are billed at provider cost on top. When customers bring their own provider keys, Vapi shows those lines as $0 on its bill, but the providers bill the customer directly. Telephony is separate and carries no per-minute figure on the reviewed Vapi pricing page. Vapi pricing, accessed 2026-08-18.

Build includes 10 concurrent calls, with added lines at $10 each per month, and retains call history for 14 days. Scale is enterprise pricing with a fixed platform fee, committed volume, volume-based per-minute rates, customizable retention, and listed SSO, RBAC, SOC 2, PCI, and data-residency options. Vapi pricing, accessed 2026-08-18.

HIPAA compliance is a $2,000 monthly add-on and Zero Data Retention is a $1,000 monthly add-on. Vapi's HIPAA docs state that the two modes are mutually exclusive. Its homepage states SOC 2, HIPAA, and PCI compliance; GDPR was not stated on the reviewed pages. Vapi pricing, HIPAA docs, and Vapi homepage, accessed 2026-08-18.

Vapi provides up to five free phone numbers for U.S. national use and documents Twilio number import. It also supports bring-your-own SIP through named integrations and a generic path for any SIP provider using gateway and authentication details; its docs warn that IP-based authentication can cause routing issues on shared infrastructure. Vapi phone-calling docs and SIP docs, accessed 2026-08-18.

Language coverage is provider-dependent, not one published platform total. The docs show separate ceilings for transcription and voice providers, with automatic language detection limited to specified multilingual or auto-detect modes. Test Suites can use a faster chat test or place a real phone call from a scripted scenario, record voice tests, and have an LLM judge the transcript against success criteria. Suites support up to 50 cases and five runs per case, tests are limited to 15 minutes, and voice test calls bill like regular calls. Vapi multilingual docs and Test Suites docs, accessed 2026-08-18.

ThunderPhone (this site): all-in engine rates

ThunderPhone publishes this page and sells a third shape: pay as you go with no subscription or seat fee, as an all-in engine rate. Spark is 2¢/minute, Bolt is 5¢/minute, and Storm is 9¢/minute, covering speech recognition, the language model, and the voice in one number. Premium voice or language paths add up to 3¢/minute, and optional add-ons carry published per-minute surcharges. Telephony is setup-dependent.

The catalog spans 47 languages and 123 voices, with automatic mid-call switching when additional languages are configured. Telephony supports SIP trunks and bring-your-own numbers through direct VoIP connections or manual SIP configuration. Testing includes AI-caller simulations, reusable graded scenarios, regression suites with minimum pass-rate gates, CI execution, and live-traffic A/B experiments; simulations are billable real calls, with the charge shown before a run. The public compliance posture is GDPR and HIPAA compliant, and customers remain responsible for configuring their workflows, data handling, recordings, integrations, and notices for their own obligations.

At a glance

The workload row uses 2,000 calls × 5 connected minutes = 10,000 monthly minutes, U.S. telephony, one number, fewer than 10 simultaneous calls, no optional add-ons, and platform voices where specified. Arithmetic is illustrative, not a quote.

Choose Retell AI if…

• You want public component prices. Infrastructure, model, voice, U.S. telephony, numbers, concurrency, and add-ons all have published figures. Retell pricing, accessed 2026-08-19.
• You need more included concurrency. Retell includes 20 calls and prices each added line at $8 per month, versus Vapi's 10 and $10. Retell pricing, accessed 2026-08-19; Vapi pricing, accessed 2026-08-18.
• A documented GDPR path matters. Retell publishes DPA and SCC details, alongside its note about no current EU operations. Retell compliance docs, accessed 2026-08-18.
• You prefer native telephony choices with a public U.S. rate. Retell publishes Twilio usage and number prices while charging nothing for custom SIP trunking. Retell pricing, accessed 2026-08-19.

Choose Vapi if…

• You already own provider accounts and keys. Vapi can leave speech, model, and voice billing directly with those providers. Vapi pricing, accessed 2026-08-18.
• You want chat and real-call tests in one suite. Vapi documents both modes, recorded voice results, LLM evaluation, and repeated runs. Vapi Test Suites docs, accessed 2026-08-18.
• A prototype benefits from included U.S. numbers. Vapi documents up to five free U.S. national numbers per account. Vapi phone-calling docs, accessed 2026-08-18.
• Your SIP provider needs a generic integration path. Vapi documents gateway and authentication details for any SIP provider, subject to its routing warning. Vapi SIP docs, accessed 2026-08-18.

Choose ThunderPhone if…

• You want the usage forecast to start from one all-in engine rate instead of a provider stack.
• You need a fixed catalog total rather than provider-specific language ceilings.
• Your release process needs pass-rate gates, CI execution, SIP-loopback tests, or live experiments.
• You want GDPR and HIPAA in the public posture without Vapi's listed monthly HIPAA add-on.

How to evaluate

1. Build the final configured minute. For Retell, add infrastructure, the exact LLM, TTS, telephony, and required add-ons. For Vapi, add hosting to the actual provider and carrier bills, including charges outside Vapi.
2. Model peak concurrency, not average volume. Price the busiest hour against 20 included Retell calls and 10 included Vapi calls, then add monthly line charges.
3. Test the telephony route. Verify number ownership, inbound and outbound directions, SIP authentication, carrier charges, and failure handling with the production path.
4. Match compliance mode to the data flow. Confirm BAAs, DPA terms, retention, provider eligibility, and whether mutually exclusive modes create a gap.
5. Run identical tests on each platform. Include tool calls, interruptions, long silence, noisy audio, and an escalation. Record real-call test charges.
6. Inspect the resulting evidence. Compare recordings, transcripts, grading criteria, repeatability, and whether the test result can block a release.

FAQ

Is Retell AI cheaper than Vapi?

The reviewed public pages do not establish a universal winner. Retell's budget configuration in the defined workload computes to $882 with one number, while Vapi computes to $500 for hosting alone and leaves speech, model, voice, and telephony outside that figure. The complete Vapi total is not publicly computable. Retell pricing, accessed 2026-08-19; Vapi pricing, accessed 2026-08-18.

Is Vapi really $0.05 per minute?

$0.05 is the Build hosting fee. STT, LLM, TTS, and telephony sit on top. Bringing keys changes who sends the provider bill, not whether the underlying services have a cost. Vapi pricing, accessed 2026-08-18.

What is Retell AI's true per-minute price?

It depends on the selected stack. Every call starts with $0.055 infrastructure, then adds the model, TTS, and telephony. Published examples range from $0.088 for the cheapest listed U.S. stack to $0.43 for GPT 5.5 fast with ElevenLabs and U.S. telephony, before a number and optional add-ons. Retell pricing, accessed 2026-08-19.

Which has better testing, Retell or Vapi?

They document different workflows. Retell publishes simulation test cases, batch runs, and mocks, plus paid AI Quality Assurance. Vapi publishes chat and real-phone Test Suites with recordings, repeated runs, and LLM judging. “Better” depends on whether pre-launch simulation, audible calls, or post-call QA is the requirement. Retell simulation docs, accessed 2026-08-18; Retell pricing, accessed 2026-08-19; Vapi Test Suites, accessed 2026-08-18.

Do Retell and Vapi both support SIP trunks?

Yes. Retell advertises no platform charge for custom telephony through SIP. Vapi documents named and generic SIP-provider paths, with a warning about IP authentication on shared infrastructure. Retell pricing, accessed 2026-08-19; Vapi SIP docs, accessed 2026-08-18.

How do their HIPAA options compare?

Retell requires a signed BAA before PHI; Vapi lists HIPAA as a $2,000 monthly add-on. Vapi also lists Zero Data Retention at $1,000 per month and says the two modes are mutually exclusive. Confirm provider configuration and agreement scope with each vendor. Retell compliance, accessed 2026-08-18; Vapi pricing and HIPAA docs, accessed 2026-08-18.

Sources and freshness

For the line-by-line rate cards, see Retell AI pricing explained and Vapi pricing explained. Related product context is available in Retell alternatives, Vapi alternatives, ThunderPhone vs. Retell AI, and ThunderPhone vs. Vapi.

Retell pricing claims were checked August 19, 2026; its language, compliance, telephony, and simulation claims retain their August 18 access date. Vapi claims were checked August 18, 2026. ThunderPhone facts were checked against its published pricing on September 10, 2026. This comparison is documentation-based: no vendor account, provider bill, invoice, or live call was tested. Recompute the stack from current provider prices before relying on it.