Vapi vs Retell vs Bland vs Synthflow in 2026: The Real All-In Per-Minute Cost of Voice AI

Three of the four advertise a per-minute rate and one has stopped publishing altogether. None of the advertised numbers is what you actually pay. Here is the honest 2026 all-in math, built only from rates the vendors publish, per sales use-case, with latency and telephony detail.

Last verified June 2026; every per-minute rate on this page re-read from the vendor's own pricing page on 23 September 2026.

$0.082-$0.129/min

Vapi all-in (own calculator)

$0.149/min

Retell mid sales stack

$0.12-$0.14/min + plan

Bland per-minute (all-in)

Not published

Synthflow

§Why Voice AI Is a Four-Component Bill

Every voice AI platform advertises a headline per-minute rate. That rate almost never reflects total cost, because a complete voice agent requires four components, and platforms expose the bundle differently. Before comparing Vapi to Retell to Bland, you need to know which components each includes and which you are bringing yourself. The rates below are the ones Vapi and Retell publish for each layer on their own pricing pages, not generic market estimates.

STT

Speech-to-Text

Deepgram on Vapi's calculator

$0.0095-$0.0099/min

LLM

Language Model

OpenAI, Anthropic, Gemini

$0.0016-$0.32/min

TTS

Text-to-Speech

ElevenLabs, Cartesia, Minimax, Fish

$0.015-$0.040/min

Telephony

Phone Network

Twilio, Telnyx, Vonage, bundled PSTN

Free-$0.015/min

§Vapi - Most Flexible Pipeline, and the Cheapest All-In

Hosting rate

$0.05/min

Plus transcriber + model + voice, at cost

+$0.032-$0.079

All-in (Vapi's own calculator)

$0.082-$0.129/min

Correction, 23 September 2026: this page previously put Vapi's all-in at $0.25 to $0.33 a minute and called it the most expensive of the group. Vapi's own calculator does not support that. It estimates $82 to $129 for a 1,000-minute month, which is $0.082 to $0.129 a minute, roughly a third of the figure carried here before. The old number assumed model and voice rates an order of magnitude above what Vapi passes through. Full working on the Vapi pricing page.

Vapi is the most developer-friendly voice AI platform with the strongest community and the most flexible pipeline, and it passes provider costs through with no markup. Its calculator prices Deepgram transcription at $0.0095 to $0.0099 a minute, an OpenAI model at $0.0077 to $0.0452, and an ElevenLabs voice at $0.0146 to $0.0238. Vapi Telephony, SIP, WebSockets and Daily WebRTC are free; Telnyx is $0.0055 a minute, Vonage $0.00814, Twilio $0.008 inbound and $0.014 outbound. The $0.05 hosting fee orchestrates all of it.

Strengths: Cheapest all-in on published rates; maximum flexibility in pipeline composition; models pass through at provider cost; free first-party telephony; strongest developer community and documentation.

Weaknesses: Support is a separate purchase (Success Packages at $29 a month for Core, a $999 a month minimum for Pro); HIPAA-eligible handling is a $2,000 a month add-on; extra concurrency is $10 per line per month; compliance setup (TCPA consent capture, GDPR recording disclosure) is entirely your responsibility.

§Retell AI - One Invoice, Four Metered Lines

Retell Voice Infra

$0.055/min

Plus TTS + model + telephony

+$0.021-$0.215

Published band / mid sales stack

$0.07-$0.31 / $0.149/min

Correction, 23 September 2026: $0.07 is the floor of a published band, not a bundle that includes a model and telephony. Retell meters four lines separately, and this page previously described the $0.07 as covering speech-to-text, text-to-speech and telephony with only the model left to add. Read from Retell's pricing page, a realistic sales configuration is $0.149 a minute. Component table on the Retell pricing page.

Retell publishes $0.07 to $0.31 a minute for voice agents and publishes the component rates behind it: Retell Voice Infra at $0.055 a minute, text-to-speech at $0.015 on six providers or $0.040 on ElevenLabs, a model line running $0.0016 to $0.32, and telephony at $0.015 through Retell's Twilio or Telnyx routing, or nothing if you bring your own SIP trunk. A budget stack lands at $0.076, a mid sales stack at $0.149 and a premium stack at $0.270. Retell reports inbound latency of 250-400ms, which is best-in-class among the four platforms.

Strengths: Best-in-class latency for inbound voice; one invoice and one vendor relationship; a published add-on menu (knowledge base, denoising, guardrails, PII removal) rather than quote-only extras; free telephony on your own SIP trunk; $10 in credits and 20 free concurrent calls to start.

Weaknesses: More expensive per minute than Vapi and Bland on published rates; no volume tiers published at all, so scale pricing is a quote; less pipeline flexibility than Vapi; compliance responsibility still falls on you.

§Bland AI + Synthflow

Bland AI

$0.12-$0.14/min all-in + $0 or $299/mo platform fee

Bland uses a platform-fee-plus-per-minute model: a free Start tier at $0.14/min and Build at $299/month and $0.12/min, with a custom-quoted Enterprise tier. The $499/month Scale tier at $0.11/min that this page previously listed is no longer on Bland's pricing page, checked 23 September 2026. The per-minute rate is all-in talk time (LLM, speech-to-text, text-to-speech, and telephony included, no separate token charges). Transfer charges and voice-cloning add-ons sit on top. Good out-of-box voice quality and simple onboarding.

Strength: the easiest bill of the four to forecast, one rate per tier with no model pass-throughs, and the only one of the four with a dialer and sequencer included. On published rates it now undercuts Retell rather than sitting above it.

Synthflow

Not published; Enterprise contracts from $30,000/year

Synthflow has stopped publishing a per-minute rate. Its pricing page now carries one line on cost, that enterprise contracts start at $30,000 annually, with everything else scoped on a call. The $0.09/min voice engine and the $0.15 to $0.24 all-in band this page previously quoted are not figures Synthflow publishes today, and they are not restated here as current. That makes Synthflow the only one of the four that cannot be modelled from published rates: at 10,000 minutes a month the $30,000 floor works out at $0.25 a minute, but that is arithmetic on a contract minimum, not a quoted rate. Synthflow positions as the simplest end-to-end option for non-developer teams; its flow builder is simpler than Vapi's but less powerful for complex conversation logic.

Strength: accessible to non-developers; no component assembly required. Weakness: no published rate, and a $30,000 annual floor that rules it out for pilots.

§Platform Comparison Table

Platform	Advertised	All-in typical	Latency	Telephony	Best for
Vapi	$0.05/min hosting	$0.082-$0.129/min	300-600ms	Vapi Telephony/SIP free; Twilio, Telnyx, Vonage priced	Cheapest all-in; maximum flexibility; developer-first
Retell AI	$0.07-$0.31/min published band	$0.076-$0.270/min ($0.149 mid)	250-400ms (best)	$0.015/min, or free on own SIP	Best latency; one invoice; published add-on menu
Bland AI	$0.14/min Start, $0.12/min Build	$0.12-$0.14/min + $0 or $299/mo	300-500ms	Included	Bundled dialer; simplest bill to forecast
Synthflow	Not published	Not published	400-600ms	Included	Non-developer teams; enterprise contracts from $30,000/yr
ElevenLabs Conversational	From $0.05/min TTS	Add STT + LLM + telephony	Improving in 2026	BYOT	Highest voice naturalness
PlayHT Conversational	$0.01-$0.04/min TTS	Add STT + LLM + telephony	Competitive	BYOT	Competitive TTS pricing
Deepgram Voice Agent	$0.0043/min STT	Add LLM + TTS + telephony	Low-latency STT	BYOT	STT-first use cases; lowest STT rate

§Sales Use-Case Sizing

Inbound qualification

Low risk

3-min avg call, 500 calls/month = 1,500 minutes

Retell

$224/mo (1,500 x $0.149 mid)

Vapi

$123-$194/mo (1,500 x $0.082-$0.129)

Bland

$210/mo (Start: 1,500 x $0.14)

Legal note: Low risk (inbound initiated)

Outbound appointment confirmation

Moderate risk

1-min avg call, 2,000 calls/month = 2,000 minutes

Retell

$298/mo (2,000 x $0.149 mid)

Vapi

$164-$258/mo (2,000 x $0.082-$0.129)

Bland

$280/mo (Start: 2,000 x $0.14)

Legal note: Moderate risk (prior relationship, not cold)

Outbound cold calling

LEGAL RISK

2-min avg call, 5,000 calls/month = 10,000 minutes

Retell

$1,490/mo (10,000 x $0.149 mid)

Vapi

$820-$1,290/mo (10,000 x $0.082-$0.129)

Bland

$1,400/mo (Start; Build is $1,499 here)

Legal note: HIGH RISK: TCPA requires prior express consent for AI voice

§Compliance Checklist

Critical: TCPA Prior Consent for Outbound AI Voice

FCC Declaratory Ruling 24-17 (adopted 8 February 2024) classified AI-generated voices as "artificial or prerecorded voice" under the TCPA. Outbound AI voice calls without prior express consent are illegal in the US. This applies to Vapi, Retell, Bland, and Synthflow when used for outbound cold calling. Prior express written consent is required for B2C marketing calls.

Full legal compliance guide →

Low

Inbound US call with AI voice agent

Lead initiated; AI disclosure required at call start; CIPA recording disclosure for California

Moderate

Outbound appointment confirmation (known customer)

Prior business relationship may support; get legal opinion on 'established business relationship' scope in your state

Illegal

Outbound AI cold call (no prior consent)

TCPA violation; prior express consent required before placing AI voice calls

CIPA

Recording any call (US, any AI)

All-party consent required for California-resident callers; AI recording may be subject to Ambriz v. Google CIPA theory

GDPR

AI recording or transcription (EU caller)

Affirmative consent required; 'this call may be recorded' notification alone is insufficient under GDPR Article 6(1)(a)

§FAQ

How much does Vapi cost per minute?

Vapi's hosting rate is $0.05 per minute, and that is not the total cost, but the rest is smaller than it once looked. Vapi passes provider costs through at cost and publishes the ranges on its own calculator: Deepgram transcription at $0.0095 to $0.0099 per minute, an OpenAI model at $0.0077 to $0.0452, and an ElevenLabs voice at $0.0146 to $0.0238, with Vapi Telephony and SIP free. Vapi's calculator adds those to $82 to $129 for a 1,000-minute month, which is $0.082 to $0.129 per minute all-in. Routing over Twilio outbound adds $0.014 per minute.

Vapi vs Retell - which is cheaper all-in?

Vapi, and this reverses what this page said before September 2026. The old answer rested on a Vapi all-in of $0.25 to $0.33 per minute and on the belief that Retell's $0.07 bundled speech-to-text, text-to-speech and telephony. Neither holds. Vapi's own calculator puts an all-in minute at $0.082 to $0.129, and Retell meters four lines separately, so a comparable Retell stack of $0.055 voice infrastructure plus $0.015 text-to-speech plus a $0.064 recommended model plus $0.015 telephony is $0.149 per minute. That makes Vapi 13 to 45 percent cheaper per minute. Retell's genuine advantages are latency, a single invoice and a published add-on menu, not price.

Is AI cold calling legal with Vapi or Retell?

Not without prior express consent. The FCC's Declaratory Ruling 24-17 (February 2024) classified AI-generated voices as artificial/prerecorded under the TCPA. Outbound AI cold calls without prior express written consent for marketing purposes are a TCPA violation. This applies to Vapi, Retell, Bland, and Synthflow. Consult qualified counsel for your specific use case before deploying outbound voice AI.

What is the latency of Retell AI?

Retell AI reports 250-400ms end-to-end latency for inbound voice qualification use cases, which is best-in-class among the major platforms. Vapi typically runs 300-600ms; Bland and Synthflow 300-600ms. Latency below 500ms is generally acceptable for natural conversation; above 600ms users start to notice pauses. Since Vapi is now the cheaper platform per minute on published rates, latency rather than cost is the strongest remaining case for Retell on inbound.

Can I build an AI cold caller with Vapi legally?

You can build it technically. Whether you can operate it legally in the US depends on your consent infrastructure. The FCC (Ruling 24-17, February 2024) requires prior express consent before placing AI voice calls to any number. For B2C marketing calls, prior express written consent is required. Without a robust consent-capture mechanism linked to every number you dial, outbound AI cold calling in the US is a TCPA violation. See the full legal guide.

TCPA + CIPA legal guide →Pipeline generation use cases →Voice AI cost calculator →