Voice AI phone agent costs in 2026: the per-minute math across four platforms
· 12 min read

Voice AI phone agent costs in 2026: the per-minute math across four platforms

Voice AI phone agent costs per connected minute across Deepgram, Vapi, Retell AI and Twilio ConversationRelay, with 1,000- and 10,000-minute examples.

A voice AI phone agent answers or places phone calls and holds a spoken conversation with the caller. Pricing pages make these tools look simple: a rate per minute. The actual bill is usually more than that one number. This guide breaks the cost into its parts, puts published rates for four platforms side by side, and shows the arithmetic so you can redo it with your own call volumes.

Thesis: The headline rate is never the whole bill. The lowest published Deepgram Voice Agent rate ($0.050 per minute) leaves the LLM and voice you bring to be paid elsewhere, and Vapi’s itemized stack adds up to $0.0898 to $0.1369 per connected minute on its published component rates.

Key terms in plain English

  • Connected minute: time the call is live with the agent talking. This is the meter most platforms bill.
  • STT (speech-to-text): turns the caller’s voice into words.
  • LLM (large language model): decides what the agent says next.
  • TTS (text-to-speech): produces the agent’s synthetic voice.
  • Transport: the carrier charge for the phone line itself.
  • Concurrency: how many calls can run at the same time.
  • BAA (business associate agreement): a HIPAA contract some regulated businesses need before handling health data.

The cost formula

cost per connected minute = platform fee + transport + STT + LLM + TTS

Per call, multiply by how long calls actually stay connected:

cost per call = cost per connected minute × average connected minutes per call

Some platforms bundle several of these into one rate. Anything not in that rate shows up somewhere else on the bill, so check every line before comparing.

Published rates, platform by platform

Deepgram Voice Agent

Deepgram’s pay-as-you-go plan includes a free $200 credit with no minimums. Its Growth plan is a prepaid annual commitment shown as $4K+ per year, with savings of up to 20% (Deepgram pricing). Voice Agent API usage is billed on websocket connection time.

List rates below come from a September 2026 third-party reprint of Deepgram’s pricing (Privocio):

  • Standard: $0.075 per minute
  • Standard BYO-TTS: $0.065 per minute
  • Custom BYO-LLM: $0.065 per minute (the reprint’s pay-as-you-go figure; the official page’s cell is ambiguous, with $0.059 appearing in one column, so confirm on the live page)
  • Custom BYO-LLM+TTS: $0.050 per minute
  • Advanced: $0.163 per minute
  • Advanced BYO-TTS: $0.122 per minute

The BYO variants list lower because you supply that component yourself. Telephony is not included, so you bring Twilio or a SIP carrier (Deepgram pricing). The entry tier allows up to 45 concurrent WSS connections. A BAA is available for Enterprise customers only. A promotional rate ran through 2026-09-12 and reverted to list on 2026-09-13 (Privocio). This guide uses list rates only.

Vapi

Vapi’s pay-as-you-go plan has a $0 platform fee, $5 in free credits, and 4 concurrent calls. The Core plan is $29 per month with 10 concurrent calls. The Pro plan starts at a $999 per month minimum (Vapi pricing).

Vapi’s hosting fee is $0.05 per minute. Its estimate prices each component per minute: STT (Deepgram transcriber) at $0.0095 to $0.0099, LLM (OpenAI) at $0.0077 to $0.0452, and TTS (ElevenLabs) at $0.0146 to $0.0238. Vapi telephony and SIP are $0. Carriers bill transport separately: Twilio inbound at $0.008 per minute, Twilio outbound at $0.014, Vonage at $0.00814, and Telnyx at $0.0055.

Adding the hosting fee and Twilio inbound transport (the inbound carrier assumption; Twilio outbound at $0.014 would add $0.006) to the component bounds gives the full stack:

  • Low: $0.05 + $0.0095 + $0.0077 + $0.0146 + $0.008 = $0.0898 per minute
  • High: $0.05 + $0.0099 + $0.0452 + $0.0238 + $0.008 = $0.1369 per minute

Extra concurrency costs $10 per line per month. A HIPAA BAA is a $2,000 per month add-on.

Retell AI

Retell’s pay-as-you-go rate for AI voice agents is $0.07 to $0.31 per minute. It includes $10 in free credits and 20 concurrent calls. Extra concurrency is $8 per concurrent call per month (Retell pricing).

The pricing page’s worked example totals $0.150 per minute: LLM $0.080, Retell voice infrastructure $0.055, TTS $0.015, and telephony $0.000. Only connected calls are billed, but voicemail is billed while the agent is active. BAA terms are not listed on the Retell pricing page, so confirm with Retell before handling protected health information.

Twilio ConversationRelay

Twilio’s conversational AI pricing page states its pricing is current as of August 2026 (Twilio conversational AI pricing). For general Twilio rates, see Twilio pricing.

A third-party aggregator lists ConversationRelay at $0.07 per minute (Thunderphone, accessed 2026-09-10). This guide does not treat that figure as verified on Twilio’s own site. Voice minutes and the LLM are billed separately. Operators are billed per character: Conversation Orchestrator at $0.0002 per 1K characters ingested, and Twilio-authored Conversation Intelligence operators at $0.005 per 1K characters. The pricing page we reviewed does not list entry-tier concurrency or BAA terms, so confirm both with Twilio.

Side-by-side comparison

PlatformPricing modelPublished per-minute rateBilled separatelyEntry-tier concurrencyBAA
Deepgram Voice AgentPay as you go, billed on websocket connection time; Growth prepaid annual from $4K+ per yearStandard $0.075; Standard BYO-TTS $0.065; Custom BYO-LLM $0.065; Custom BYO-LLM+TTS $0.050; Advanced $0.163; Advanced BYO-TTS $0.122Telephony (not included); the LLM and TTS you bringUp to 45 WSS connections on PAYGEnterprise only
VapiPay as you go ($0 platform fee, $5 credits); Core $29/mo; Pro from $999/moHosting $0.05; STT $0.0095–$0.0099; LLM $0.0077–$0.0452; TTS $0.0146–$0.0238; Vapi telephony/SIP $0Carrier transport: Twilio inbound $0.008, outbound $0.014, Telnyx $0.0055, Vonage $0.008144 on PAYG; 10 on Core; extra $10 per line per monthHIPAA add-on $2,000/mo
Retell AIPay as you go, $10 free credits$0.07–$0.31 range; page example $0.150 (LLM $0.080, voice infra $0.055, TTS $0.015, telephony $0.000)Not itemized on the pricing page beyond the example; billed on connected calls20 included; extra $8 per concurrent call per monthNot listed on pages reviewed
Twilio ConversationRelayPer-minute relay plus separately billed voice, LLM and operators$0.07 per minute per Thunderphone (not verified on twilio.com)Voice minutes; LLM; Orchestrator $0.0002 per 1K characters; Intelligence operators $0.005 per 1K charactersNot listed on pages reviewedNot listed on pages reviewed

Worked example: 1,000 and 10,000 connected minutes

The table below multiplies published rates by connected minutes. It includes only the components named in each row.

Platform and basisPer-minute input1,000 connected minutes10,000 connected minutes
Deepgram Standard, list$0.075$75$750
Deepgram Custom BYO-LLM+TTS, list$0.050$50$500
Vapi stack, with Twilio inbound$0.0898–$0.1369$89.80–$136.90$898–$1,369
Retell AI, full published range$0.07–$0.31$70–$310$700–$3,100
Retell AI, page example$0.150$150$1,500
Twilio ConversationRelay, relay only$0.07$70$700

Read these rows carefully:

  • Twilio’s $0.07/min worked-example row is relay-only and excludes voice minutes, LLM, and operator character fees. Operator fees depend on character volume, which this guide does not estimate, so no full Twilio total is given.
  • The Deepgram rows exclude telephony and any external LLM or TTS you pay for separately. The BYO-LLM+TTS row is only a complete cost if you already have those providers priced in.
  • The Vapi row assumes the example components (Deepgram transcriber, OpenAI, ElevenLabs) and Twilio inbound transport. Choosing other carriers or models changes it.

Deployment advice

Plan concurrency before choosing a tier

Concurrency limits are often the first constraint you hit, not the price. Vapi’s pay-as-you-go plan allows 4 concurrent calls, so any setup that rings more lines at once needs Core ($29 per month, 10 concurrent) or paid extra lines at $10 per line per month. Retell includes 20 concurrent calls and charges $8 per concurrent call per month above that. Deepgram’s entry tier allows up to 45 WSS connections. The Twilio pages we reviewed do not list entry-tier concurrency, so check it before you commit.

Budget at list rates, not promotions

Deepgram’s promotional rate ended on 2026-09-13. Blog posts and screenshots from the promotional period will overstate how cheap the product looks today. Build your budget from the list rates, and recheck any pricing page before you sign up.

Decide who bills your telephony

Telephony is the line item most often left out of a quick estimate. Deepgram does not include it, so you choose a carrier or SIP provider and pay them directly. Vapi’s telephony is $0, but carrier transport is charged on top: Twilio inbound is $0.008 per minute and Telnyx is $0.0055 per minute on Vapi’s list. Retell’s example shows telephony at $0.000, but that is one example, not a general promise. Read the carrier line on each invoice.

Weigh BYO-LLM and BYO-TTS trade-offs

Bring-your-own options lower the platform rate. Deepgram’s Standard BYO-TTS list rate is $0.010 per minute below Standard, and BYO-LLM+TTS is $0.025 per minute below Standard. The saving is only real if your external LLM and TTS bills come in lower than the bundled component, so price those providers separately. Vapi’s estimate shows the same logic in reverse: each component is a separate meter, and each one can be swapped. Keep components separate in your cost tracking so you can change one without guessing what it did to the total.

Failure modes and production measurement

  • Idle connections billing. Deepgram bills on websocket connection time, so a socket left open after a call ends keeps costing money. Confirm that your application closes sockets on hang-up.
  • Voicemail billing. Retell bills voicemail while the agent is active. Decide how your flow handles machines before you run campaigns.
  • Concurrency ceilings. A plan’s concurrency cap is a planning constraint. Size peak simultaneous calls against the tier before you launch.
  • Promotional rates. Budgets built from expired discounts will undercount. Use list rates.
  • Unmodeled meters. Twilio operator fees are priced per character, so a cost model built only on per-minute rates will miss them.

For measurement, reconcile each platform’s invoice against your own connected-minute logs every billing cycle. Track cost per completed call, not just cost per minute, and watch peak concurrency against your plan. Set a budget alert on each platform so a bad loop or a leaked socket shows up within days, not at month end. This guide does not make claims about call quality, latency, or accuracy; measure those separately with your own call data.

Frequently asked questions

Is the lowest per-minute rate the cheapest bill?

Not necessarily. Deepgram’s lowest published rate, Custom BYO-LLM+TTS at $0.050 per minute, excludes the LLM and TTS you supply and excludes telephony. Vapi’s itemized stack computes to $0.0898 to $0.1369 per minute on published component rates. Compare complete stacks, not headline rates.

Which platforms list a BAA?

Deepgram offers a BAA for Enterprise customers only. Vapi lists a HIPAA BAA add-on at $2,000 per month. BAA availability for Retell AI and Twilio ConversationRelay is not listed on the pricing pages reviewed, so confirm with each vendor before handling protected health information.

How many calls can I run at once on the entry tier?

Deepgram’s pay-as-you-go tier allows up to 45 concurrent WSS connections. Vapi’s pay-as-you-go tier allows 4 concurrent calls, and Core allows 10. Retell includes 20 concurrent calls. Twilio’s entry-tier concurrency is not listed on the pricing pages reviewed.

Is Twilio ConversationRelay’s $0.07 per minute the full cost?

Not on its own. The $0.07 per minute figure comes from a third-party aggregator and is reported to cover relay only. Voice minutes and the LLM are billed separately, and operator fees are charged per 1K characters at $0.0002 (Orchestrator) and $0.005 (Intelligence operators). This guide does not treat the $0.07 figure as verified on Twilio’s own site.

The Bottom Line

No platform in this comparison wins on every line. Deepgram’s lowest list rate ($0.050, Custom BYO-LLM+TTS) excludes the LLM and TTS you supply, and telephony is not included and a BAA requires Enterprise. Vapi itemizes every component, which makes its costs easy to trace, but the full stack is higher than its headline hosting fee suggests. Retell publishes a wide per-minute range and a clear concurrency add-on, but does not list BAA terms on its pricing page. Twilio ConversationRelay’s quoted per-minute figure is relay-only, so its full cost depends on the separate meters.

Choose by the complete stack for your use case, not by the headline rate. Keep every component meter separate, so you can swap a model, a voice, or a carrier without losing track of the total. Avoid tying your monthly budget to a promotional rate. This guide does not score the platforms, because published pricing alone cannot support a quality ranking.

How This Guide Was Built

This review is based on official pricing pages, third-party reprints, and arithmetic on published rates. We did not run any of these platforms hands-on.

Sources and fetch dates:

Method: Every per-minute and per-call figure is a published rate or simple arithmetic on published rates. Totals show their formulas. Promotional Deepgram rates are excluded as expired.

Not covered: This guide makes no claims about latency, word error rate, accuracy, or call quality. It reports no hands-on testing, benchmarks, or production results.

  • ToolBrain — tool reviews, LLM comparisons, and AI workflow guides
  • CodeIntel Log — code quality, debugging, and software engineering benchmarks

Cross-links automatically generated from None.

Back to all posts