Quick answer: Vapi charges $0.05 a minute to orchestrate the speech-to-text, model and voice you pick, each billed at provider prices, and includes a free inbound US number. Deepgram’s Voice Agent API bundles all three into one rate ($0.075 Standard, $0.050 if you bring your own LLM and voice) but sells no phone line. At 10,000 inbound minutes we itemise $741.60–$1,267.25 on Vapi and $776.45–$1,760.15 on Deepgram with Twilio.
“Vapi vs Deepgram” is a choice between two ways of buying the same three parts of a voice agent: speech-to-text, a large language model (LLM) and text-to-speech. Vapi is an orchestration platform. It charges a hosting fee per minute and passes through whichever providers you pick, Deepgram included. The Deepgram Voice Agent API is one vendor’s bundle: a single WebSocket that runs Deepgram speech-to-text, a managed LLM and Deepgram or Cartesia text-to-speech, plus turn-taking and barge-in, for one per-minute rate. What it does not include is a phone line. To answer a call you bridge Twilio, Plivo, Amazon Connect, Genesys or AudioCodes audio into it yourself. Every price below was read from vapi.ai, docs.vapi.ai, deepgram.com, developers.deepgram.com and Twilio’s own US voice pricing page on 20 November 2026, in US dollars.
Where Zian fits: an AI agents platform, not a developer API
If the job is calls answered and appointments booked, rather than a voice pipeline to assemble and maintain, Zian AI is the platform to look at first, and LeadsNow.ai is the done-for-you route. Zian AI is a standalone AI sales and reception agents platform, not a phone system (carrier / CPaaS / contact centre) and not an add-on to one.
Zian answers inbound calls 24/7 as an AI receptionist that qualifies, books, transfers and takes messages, runs outbound calls, and follows up by SMS, email, WhatsApp, Instagram DMs and Facebook Messenger, with answering-machine and voicemail handling on calls. SmartReach AI™ orchestrates the message, channel and timing of follow-up, and PrecisionPitch AI™ runs continuous split-testing of scripts and approaches. It supports 30+ languages and voice cloning, connects to HubSpot, Salesforce, HighLevel and Zapier, and offers private model deployment on customer infrastructure. Zian has set 64,521+ qualified sales appointments. LeadsNow.ai runs on Zian as a managed, pay-per-result service that answers leads in under 60 seconds, 24/7, and has set 50,769+ appointments. Zian is in a partnership-application beta with no published prices; each of its three listed plans, including Intel + Widgets (Free Forever), starts with a partnership application.
Apply For Partnership to see whether Zian’s inbound and outbound agents fit your call volume.
Vapi vs Deepgram Voice Agent API: what each one actually is
Deepgram’s architecture page describes its product as one API “for the entire voice conversation”: “You open one WebSocket connection, send audio in, and receive audio out.” Inside, it is still a cascade of three stages, which Deepgram’s configuration calls listen, think and speak. Listen is Deepgram only. Think can be a Deepgram-managed OpenAI, Anthropic, Google or NVIDIA model, your own endpoint, or an Amazon Bedrock or Groq deployment you supply. Speak defaults to Deepgram’s Flux TTS and can instead be Aura, Deepgram-managed Cartesia, or your own OpenAI, ElevenLabs, Cartesia or Amazon Polly account.
Vapi’s pricing FAQ describes its per-minute cost as “Four components: speech-to-text, the language model, text-to-speech, and transport. Each is billed at the provider’s listed price with no markup. Vapi’s $0.05/min hosting fee is added on top.” Vapi’s FAQ adds that “You can bring your own custom models for any part of the pipeline.” Deepgram is one of the providers you can pick inside Vapi, for transcription (Nova and Flux) and for Aura voices.
| Question | Vapi | Deepgram Voice Agent API |
|---|---|---|
| What you buy | Hosting and orchestration at $0.05 a minute, plus each provider at its listed price | One per-minute rate covering speech-to-text, LLM and text-to-speech, in six published tiers |
| Speech-to-text | Your pick of providers (calculator lists Deepgram, AssemblyAI, OpenAI); Deepgram Flux available | Deepgram only: Flux (model-integrated end of turn) or Nova (voice activity detection) |
| Language model | Your pick (calculator lists OpenAI, Anthropic, Google), or a custom LLM URL | Deepgram-managed OpenAI, Anthropic, Google, NVIDIA models in a Standard or Advanced tier, or bring your own |
| Text-to-speech | Your pick (calculator lists ElevenLabs, Vapi voices, Cartesia); Deepgram Aura and Aura-2 | Flux TTS (default), Aura, managed Cartesia, or your own OpenAI, ElevenLabs, Cartesia or Amazon Polly |
| Phone numbers | 1 free US number (inbound only) on Usage only; import Twilio or carrier numbers for outbound | None sold; guides bridge Twilio, Plivo, Genesys, Amazon Connect or AudioCodes |
| What you host | A server URL for tool calls (Vapi’s FAQ: needed “for more advanced features like function calling”) | A bridge server that relays call audio between your carrier and the WebSocket, plus your function handlers |
| Concurrent calls before you pay more | 4 on Usage only; $10 per extra line a month; Core $29 a month for 10 | Up to 45 Voice Agent connections on Pay as you go in every region; Growth up to 60 in North America |
| Self-hosting | Vapi’s FAQ: “Vapi does not support on-premise deployments” | Helm chart on your own Kubernetes cluster; self-hosted containers are an Enterprise option |
| HIPAA | $2,000 a month add-on with a BAA | BAAs signed for Enterprise customers |
In one sentence: Vapi sells orchestration and lets you shop for every part, including the number; Deepgram sells the whole conversation loop as one metered product and leaves the phone line to you.
What Deepgram’s per-minute tiers include, and what Vapi’s $0.05 does not
Deepgram prints six Voice Agent tiers on its pricing page, “Calculated on WebSocket connection time.” The page’s structured data carries the same figures as the visible table. Its per-hour toggle rounds differently: Advanced shows $9.75 an hour, or $0.1625 a minute against $0.163, which would take $5.00 off route E below; the bill uses the per-minute figures. Which managed model lands in Standard or Advanced is set per model on Deepgram’s LLM models page.
| Deepgram Voice Agent tier | Pay as you go | Growth | What it includes, in Deepgram’s words | What you pay someone else |
|---|---|---|---|---|
| Standard | $0.075 / min | $0.068 / min | “Deepgram speech-to-text, a managed language model, and Deepgram text-to-speech, behind one connection.” Standard models include gpt-4o-mini, gpt-4.1-mini, gpt-5-mini and claude-haiku-4-5; managed Cartesia is also in Standard | Carrier charges (on Twilio: minute, number, Media Streams), your bridge server |
| Standard, BYO TTS | $0.056 / min | $0.051 / min | “The Standard tier with your own text-to-speech provider in place of ours.” | Your voice vendor, plus the above |
| Custom, BYO LLM | $0.065 / min | $0.059 / min | “Your own language model, with Deepgram speech-to-text and text-to-speech.” | Your LLM vendor, plus the above |
| Custom, BYO LLM and TTS | $0.050 / min | $0.040 / min | “Your own language model and text-to-speech, with Deepgram speech-to-text.” | Your LLM and voice vendors, plus the above |
| Advanced | $0.163 / min | $0.146 / min | “A higher-capability managed language model, with Deepgram speech-to-text and text-to-speech.” Advanced models include gpt-4o, gpt-4.1, gpt-5 and claude-sonnet-5 | Carrier charges (on Twilio: minute, number, Media Streams), your bridge server |
| Advanced, BYO TTS | $0.122 / min | $0.110 / min | “The Advanced tier with your own text-to-speech provider in place of ours.” | Your voice vendor, plus the above |
Growth is not a monthly plan: Deepgram says it “starts at $4,000 a year in pre-paid annual credits for up to 20% off and higher streaming concurrency”, and usage beyond the credits bills at the Growth rate plus 10%. Pay as you go starts with a $200 credit.
Vapi’s $0.05 hosting fee buys none of the three models. Its calculator bands, which Vapi says are provider prices with no markup, are Deepgram $0.00953–$0.00994, OpenAI $0.00774–$0.04523 and Cartesia $0.00689–$0.01294 a minute, as at 20 November 2026. What costs nothing on top of the $0.05 is transport on a Vapi number: Vapi lists “Vapi Telephony / SIP” as free and Twilio inbound at $0.008 a minute, “Charged by provider, not Vapi” (Twilio’s own US page shows $0.0085 for a local number, the figure our bill uses). Usage only also comes with one Vapi phone number and 4 concurrent calls.
How does a Deepgram voice agent answer a phone call?
Through a carrier you bring and a small server you run. Deepgram’s architecture page says audio comes from “a browser, a mobile app, or a phone call over your telephony provider”, and its telephony guides cover Twilio, Plivo, Genesys Cloud CX, Amazon Connect and AudioCodes LiveHub. We found no Deepgram phone number or SIP product in its documentation index, pricing page or the last 90 days of its changelog (19 August to 17 November 2026), as at 20 November 2026.
Deepgram’s Twilio guide shows the shape. You point a Twilio number at TwiML containing <Connect><Stream>, which opens a bidirectional Media Stream to your server. Your server forwards the caller’s 8 kHz mulaw audio into the Voice Agent WebSocket and plays the agent’s audio back to Twilio. For barge-in, your one job is to relay Deepgram’s UserStartedSpeaking event as a Twilio clear message. The guide’s prerequisites are “A Twilio account with a voice-capable phone number” and a Deepgram API key, and it adds: “You don’t need a separate LLM provider key.” For outbound, Deepgram’s reference implementation has your server place the call through the Twilio REST API and uses Twilio’s answering machine detection.
Two costs follow from that design. Twilio bills the call minute and the Media Stream ($0.0085 and $0.0044 a minute on a US local number, plus $1.15 a month for the number, as at 20 November 2026), and Twilio’s help centre says partial minutes are “rounded up to the next full minute” on Programmable Voice and Media Streams. And your bridge server must stay up for every call. Our Vapi vs Twilio cost breakdown prices that Twilio Media Streams route on its own, with your own models instead of Deepgram’s bundle.
On Vapi, a free Vapi number answers inbound US calls with no carrier account, but “Free Vapi numbers are inbound only. Import a number from a supported telephony provider to make outbound calls.” Importing a Twilio number takes your Account SID and Auth Token.
What does a 10,000-minute month cost on Vapi vs Deepgram?
This is the worked bill. Every line is a published rate times a quantity you can change.
Assumptions, labelled. A1: 10,000 billed minutes a month of inbound calls to one US local number (a free Vapi number cannot dial out). A2: a peak of 4 simultaneous calls or fewer, so no extra Vapi lines. A3: the same three vendors everywhere: Deepgram for speech-to-text, an OpenAI model and a Cartesia voice. On Vapi those lines use Vapi’s published calculator bands. A4: in Deepgram’s BYO column, your own OpenAI and Cartesia bills are shown at the same Vapi bands for illustration; Deepgram publishes no figure for them, and your invoices will differ. A5: Deepgram’s Standard and Advanced columns use a managed OpenAI model in that tier and a Deepgram or managed Cartesia voice, so the per-minute rate is the whole model bill. A6: Pay as you go on Deepgram, Usage only on Vapi, no HIPAA, no recording, no volume discounts. A7: the low column pairs the low end of every band, the high column the high end.
| Line item, 10,000 inbound minutes | A. Vapi, free Vapi number | B. Vapi, imported Twilio number | C. Deepgram Standard, Twilio | D. Deepgram Custom BYO LLM and TTS, Twilio | E. Deepgram Advanced, Twilio |
|---|---|---|---|---|---|
| Platform or agent fee | 10,000 × $0.05 = $500.00 | 10,000 × $0.05 = $500.00 | 10,000 × $0.075 = $750.00 | 10,000 × $0.050 = $500.00 | 10,000 × $0.163 = $1,630.00 |
| Speech-to-text (Deepgram) | $95.30–$99.40 | $95.30–$99.40 | Included | Included | Included |
| Language model (OpenAI) | $77.40–$452.30 | $77.40–$452.30 | Included (Standard model) | $77.40–$452.30, paid to OpenAI (A4) | Included (Advanced model) |
| Voice (Cartesia) | $68.90–$129.40 | $68.90–$129.40 | Included | $68.90–$129.40, paid to Cartesia (A4) | Included |
| Twilio Media Streams ($0.0044) | $0.00 | $0.00 (none listed by Vapi) | $44.00 | $44.00 | $44.00 |
| Carrier minutes (Twilio local inbound $0.0085) | $0.00 (Vapi Telephony listed as free) | $85.00 | $85.00 | $85.00 | $85.00 |
| Phone number | $0.00 (1 included) | $1.15 | $1.15 | $1.15 | $1.15 |
| Monthly total, published prices | $741.60–$1,181.10 | $827.75–$1,267.25 | $880.15 | $776.45–$1,211.85 | $1,760.15 |
| Effective cost per minute | $0.0742–$0.1181 | $0.0828–$0.1267 | $0.0880 | $0.0776–$0.1212 | $0.1760 |
| Not priced: what you run yourself | Tool-call server | Tool-call server | Bridge server and function handlers | Bridge server, function handlers, LLM and voice accounts | Bridge server and function handlers |
How we checked it. Once line by line, and once by summing per-minute rates first. Route A low: ($0.05 + $0.00953 + $0.00774 + $0.00689) × 10,000 = $741.60. Route B high: ($0.05 + $0.00994 + $0.04523 + $0.01294 + $0.0085) × 10,000 + $1.15 = $1,267.25. Route C: ($0.075 + $0.0044 + $0.0085) × 10,000 + $1.15 = $880.15. Route D low: ($0.05 + $0.00774 + $0.00689 + $0.0044 + $0.0085) × 10,000 + $1.15 = $776.45. Route E: ($0.163 + $0.0044 + $0.0085) × 10,000 + $1.15 = $1,760.15. Both methods agree to the cent in every cell. On Deepgram Growth, Standard at $0.068 makes route C $810.15 a month, but only if the year’s Deepgram usage is prepaid (about $8,160 at 10,000 minutes a month): Growth starts at $4,000 a year in credits, and usage beyond the credits bills at the Growth rate plus 10%.
What the bill says. On published prices, Vapi on its own free number is the lowest column at the low end ($741.60), and Deepgram Standard on Twilio ($880.15) sits inside Vapi’s range on both numbers. Which one wins depends on the model and voice you would run on Vapi; the rule below turns that into one number. Deepgram’s BYO column costs $51.30–$55.40 less than Vapi on the same imported Twilio number, because both charge five cents a minute but Deepgram’s five cents includes speech-to-text while Media Streams adds only $0.0044. Against a free Vapi number, which carries the call at no charge, the same BYO column costs $30.75–$34.85 more. Deepgram Advanced is the highest column at $1,760.15.
The Two-Cent Headroom Rule: when Deepgram’s bundle beats Vapi’s fee
The Two-Cent Headroom Rule: on the same Twilio number, add up what your language model and voice cost per minute on Vapi. If the sum is more than $0.0195, Deepgram’s Standard tier costs less for that call; if it is less, Vapi does. The rule holds only when your model and voice are ones Deepgram’s Standard tier covers.
The $0.0195 is the gap that is left once the lines that cancel are removed. Deepgram Standard costs $0.075 plus $0.0044 for Media Streams, or $0.0794 a minute. Vapi costs $0.05 plus Deepgram transcription at $0.00953–$0.00994, before the model and voice. Both pay the same Twilio minute and number. $0.0794 − $0.05 − $0.00994 leaves $0.01946 a minute of headroom ($0.01987 at the low end of the transcription band). Because every term is per minute, the rule holds at 2,000 minutes or 200,000.
| Comparison (per minute) | Headroom for your Vapi model + voice | Example Vapi model + voice (calculator low ends) | Which costs less |
|---|---|---|---|
| Deepgram Standard vs Vapi, both on a Twilio number | $0.01946–$0.01987 | OpenAI $0.00774 + Cartesia $0.00689 = $0.01463 | Vapi, by $48.30–$52.40 at 10,000 minutes |
| Deepgram Standard vs Vapi, both on a Twilio number | $0.01946–$0.01987 | OpenAI $0.00774 + ElevenLabs $0.0146 = $0.02234 | Deepgram Standard |
| Deepgram Standard on Twilio vs Vapi on its free number (inbound only) | $0.02796–$0.02837, plus $1.15 a month | OpenAI $0.00774 + ElevenLabs $0.0146 = $0.02234 | Vapi |
| Deepgram Advanced vs Vapi, both on a Twilio number | $0.10746–$0.10787 | OpenAI $0.04523 + Cartesia $0.01294 = $0.05817 (calculator high ends) | Vapi, across the whole OpenAI and Cartesia bands |
| Deepgram BYO LLM and TTS vs Vapi, both on a Twilio number | Not needed: model and voice cost the same on both | Any | Deepgram, by $0.00513–$0.00554 a minute |
Two adjustments move the line. Above 4 simultaneous calls, Vapi charges $10 a line a month on Usage only or $29 a month for Core’s 10 lines, while Deepgram’s Pay as you go allows up to 45 Voice Agent connections; at 10,000 minutes, Core’s $29 alone adds $0.0029 a minute to the Vapi side. And on Deepgram you run the bridge server, which this rule does not price. The rule also assumes, as the bill does, that a Vapi call on an imported Twilio number adds no Twilio Media Streams charge; Vapi’s import guide does not say either way, and if your Twilio invoice shows $0.0044 a minute there too, the line falls to about $0.0151. Vapi’s calculator does not say which OpenAI model sits at each end of its band, so read your own model’s figure from Vapi’s cost estimate before you apply the rule.
Turn-taking, function calling and languages: where the two differ
Turn-taking. Flux is not exclusive to Deepgram’s own agent. Vapi’s speech configuration lists “Deepgram Flux” as a transcriber with built-in turn detection, alongside Assembly’s end-of-turn flag, LiveKit’s model, Vapi’s own and a custom endpointing model on your server, and its Start Speaking Plan waits 0.4 seconds by default. Inside Deepgram’s Voice Agent API, Flux is the listen model in its Twilio guide, and Deepgram’s product page says Flux “detects the end of the turn in about 260 ms”. Since 28 August 2026 the agent also accepts ForceEndTurn. Our guide to diagnosing Flux cutting callers off shows how to tell a model ending from a timeout.
Function calling. Deepgram supports client-side functions, which your application runs after a FunctionCallRequest, and server-side functions, where “Deepgram calls a web endpoint that you provide.” By default a function call dispatches before the caller’s turn is confirmed; since 8 September 2026 defer_until_eot holds a function such as end_call until it is. Vapi ships built-in tools for the call itself (transfer, end call, SMS, keypad digits, voicemail), plus API Request, webhook Function and MCP tools. On Deepgram those phone actions are yours to build against your carrier: its Plivo guide gives the agent client-side transfer, SMS and callback functions, and its Think Live page says “Deepgram does not transfer the call for you.”
Languages. On Vapi, languages depend on the providers you choose; its Deepgram voice page lists Aura in English and Aura-2 in English, Spanish and German. We could not find Flux TTS on Vapi’s Deepgram voice page, docs index or changelog, as at 20 November 2026. Deepgram’s agent can speak Flux TTS in English and 41 other languages, and on 16 November 2026 Deepgram added 36 Nova-3 transcription languages. Our Deepgram voice agent language matrix counts what the agent can hear and say, language by language.
Region and session limits. Deepgram’s Voice Agent runs on US, EU, Australian and Indian endpoints. Its sessions “close automatically after 2 hours”, with a warning 5 minutes ahead.
Which one should you pick?
The table puts the platform answer first, then the build routes. It is a fit guide, not a scored ranking: the Vapi and Deepgram rows come from their own pages, and the Zian and LeadsNow.ai rows from what those companies publish. zian.ai publishes this page and appears in it.
| If you are… | Pick | Why | Wrong for you if… |
|---|---|---|---|
| A business that wants calls answered and appointments booked, not a voice pipeline | Zian AI (platform); LeadsNow.ai for done-for-you | AI receptionist and outbound agents with SMS, email, WhatsApp, Instagram DM and Messenger follow-up in one standalone platform; LeadsNow.ai runs it for you, pay-per-result | You need to write your own agent code or need a published price today |
| A developer team that wants phone numbers, transfers and outbound campaigns handled | Vapi | Free inbound number, imported carrier numbers, built-in transfer, hang-up, SMS and voicemail tools, any model mix | You must run on premises, or your model and voice cost more than the $0.0195 headroom on a Twilio number |
| A team that wants one vendor and one bill for speech and a mid-sized model | Deepgram Standard | $0.075 covers Flux, a Standard-tier model and a Deepgram or Cartesia voice; 45 connections on Pay as you go | You cannot run a bridge server, or you need an Advanced-tier model |
| A team with its own LLM contract and voice vendor | Deepgram Custom BYO LLM and TTS | $0.050 a minute includes Deepgram speech-to-text, so it undercuts Vapi on the same Twilio number | The inbound-only job could run on a free Vapi number for less |
| A team that needs speech processed in its own environment | Deepgram self-hosted (Enterprise) | Helm chart on your own Kubernetes cluster | You have no Enterprise agreement |
If neither a hosted platform nor a bundled API fits, our Vapi vs Pipecat comparison prices the open-source route on the same Deepgram, OpenAI and Cartesia models.
Rather not assemble a voice agent at all? Zian runs AI receptionist, outbound calling and messaging agents as one platform. Apply For Partnership
The sentence on each vendor’s pages that cuts against this page
On Deepgram’s side it is the product page’s “Audio never round-trips through your infrastructure.” That holds inside the API, but on a phone call Deepgram’s own Twilio guide routes every audio frame through your bridge server, and the Voice Agent rate stops at the WebSocket. Deepgram’s own LLMs are not an escape hatch either; its LLM page says they are “available only to current Deepgram contract customers enrolled in the Early Access program” and “Do not run production traffic on it.”
On Vapi’s side it is the billing guide’s “Every minute your assistants run is billed. Usage only does not include a response or uptime service-level agreement.” The $0.05 is charged whatever model you run, a response SLA starts with Core, and a 99% uptime SLA with Pro at 10% of hosting fees with a $999 a month minimum. Vapi’s FAQ does say “volume pricing is available as you scale”, without a published figure.
What this bill leaves out
The worked bill prices published per-minute lines and nothing else. It leaves out your bridge and tool servers, engineering time, taxes, recording, answering machine detection (Twilio lists $0.0075 a call), HIPAA add-ons, Vapi concurrency above 4 lines, and any committed-use or volume pricing. We could not find a billing increment for Deepgram’s Voice Agent connection minutes or for Vapi’s hosting fee, as at 20 November 2026; Deepgram’s per-second statement on its pricing page is written about audio, while the Voice Agent table says only that it is calculated on connection time. Vapi’s model bands are Vapi’s estimates, built on 5 model requests per turn, 150 output tokens a minute and a 50% prompt-cache hit rate; its transcription estimate counts “two audio channels”, and it says long calls often cost more. Vapi’s import guide does not say whether a call on an imported Twilio number adds any Twilio feature charge beyond the minute, as at 20 November 2026; check your first Twilio invoice.
Frequently asked questions
Is Deepgram’s Voice Agent API cheaper than Vapi?
It depends on the model and voice. At 10,000 inbound US minutes we itemise $880.15 on Deepgram Standard with Twilio, against $741.60–$1,181.10 on Vapi with its free number and $827.75–$1,267.25 on an imported Twilio number. Deepgram’s $0.050 BYO tier undercuts Vapi on the same Twilio number.
Does Deepgram’s Voice Agent API come with phone numbers?
No. Deepgram’s guides bridge a carrier such as Twilio or Plivo into the Voice Agent WebSocket through a server you run. The Twilio guide uses a bidirectional Media Stream, which Twilio bills at $0.0044 a minute on top of the call minute.
What does Deepgram’s Voice Agent Standard tier include?
Deepgram’s pricing page lists Standard at $0.075 a minute on Pay as you go, covering Deepgram speech-to-text, a managed language model and Deepgram text-to-speech behind one connection. Models such as gpt-4o-mini and claude-haiku-4-5 are Standard; gpt-4o and claude-sonnet-5 are Advanced at $0.163.
Can Vapi use Deepgram Flux?
Yes. Vapi’s Deepgram transcriber page lists flux-general-en and flux-general-multi, and its speech configuration lets Flux’s built-in turn detection decide when the caller has finished. On Vapi, Deepgram transcription is billed at $0.00953–$0.00994 a minute in its calculator.
Can I bring my own LLM to Deepgram’s Voice Agent API?
Yes. Deepgram accepts your own OpenAI-compatible endpoint, Anthropic or Google endpoint, Amazon Bedrock or Groq, with an ordered fallback list. The Custom BYO LLM tier is $0.065 a minute, or $0.050 if you also bring the voice, and you pay the LLM vendor yourself.
What is the Two-Cent Headroom Rule?
On the same Twilio number, Deepgram Standard costs $0.0794 a minute with Media Streams, and Vapi costs $0.05 plus about $0.0099 for Deepgram transcription. If your Vapi model and voice together cost more than $0.0195 a minute, Deepgram Standard costs less; below that, Vapi does.
How many calls can run at once on Vapi and on Deepgram?
Vapi’s Usage only includes 4 concurrent calls, with extra lines at $10 a month each or 10 on Core at $29 a month. Deepgram’s Pay as you go allows up to 45 Voice Agent connections in every region, and Growth up to 60 in North America.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Voice Agent tiers: $0.075 / $0.056 / $0.065 / $0.050 / $0.163 / $0.122 (Pay as you go), $0.068 / $0.051 / $0.059 / $0.040 / $0.146 / $0.110 (Growth); tier descriptions; “Calculated on WebSocket connection time”; Growth $4,000+ a year, rate plus 10% overage; $200 credit; BAA for Enterprise; self-hosted on Enterprise | Deepgram | deepgram.com/pricing | 20 November 2026 |
| Billing per minute of connection time; “as low as $0.050/min”; Helm chart self-hosting; Flux end of turn “about 260 ms”; “Audio never round-trips through your infrastructure” | Deepgram | deepgram.com/product/voice-agent-api | 20 November 2026 |
| Managed model tiers (Standard / Advanced); seven provider types; BYO endpoints; fallback list; Deepgram LLMs Early Access | Deepgram | developers.deepgram.com/docs/voice-agent-llm-models | 20 November 2026 |
| Flux TTS default, English and 41 other languages; managed Cartesia in Standard; BYO OpenAI, ElevenLabs, Cartesia, Amazon Polly | Deepgram | developers.deepgram.com/docs/voice-agent-tts-models | 20 November 2026 |
| Twilio bridge via <Connect><Stream>; barge-in via clear; no separate LLM key | Deepgram | developers.deepgram.com/docs/twilio-and-deepgram-voice-agent | 20 November 2026 |
| Outbound via Twilio REST API, Twilio async AMD, Aura-2 voicemail | Deepgram | developers.deepgram.com/docs/outbound-telephony-agent | 20 November 2026 |
| Telephony providers (Twilio, Plivo, Genesys, Amazon Connect, AudioCodes); “over your telephony provider” | Deepgram | developers.deepgram.com/docs/voice-agent-architecture | 20 November 2026 |
| Client-side and server-side functions; defer_until_eot | Deepgram | developers.deepgram.com/docs/voice-agents-function-calling | 20 November 2026 |
| “Deepgram does not transfer the call for you” | Deepgram | developers.deepgram.com/docs/voice-agent-think-live | 20 November 2026 |
| Voice Agent concurrency: 45 (Pay as you go, all regions); 60 North America on Growth | Deepgram | developers.deepgram.com/reference/api-rate-limits | 20 November 2026 |
| Sessions close after 2 hours | Deepgram | developers.deepgram.com/docs/voice-agent-feature-overview | 20 November 2026 |
| ForceEndTurn (28 August 2026); defer_until_eot (8 September 2026); 36 Nova-3 languages (16 November 2026) | Deepgram | developers.deepgram.com/changelog | 20 November 2026 |
| $0.05/min hosting; four components at provider price, no markup; calculator bands Deepgram $0.00953–$0.00994, OpenAI $0.00774–$0.04523, Cartesia $0.00689–$0.01294, ElevenLabs $0.0146–$0.02383; transport (Vapi Telephony free, Twilio $0.008 in); Usage only 4 calls and 1 number; Core $29 / 10 calls; Pro 10% of hosting, $999 minimum; $10 a line; HIPAA $2,000 | Vapi | vapi.ai/pricing | 20 November 2026 |
| Free Vapi numbers inbound only, US area codes | Vapi | docs.vapi.ai/free-telephony | 20 November 2026 |
| Import a Twilio number with Account SID and Auth Token | Vapi | docs.vapi.ai/phone-numbers/import-twilio | 20 November 2026 |
| “Every minute your assistants run is billed”; no SLA on Usage only | Vapi | docs.vapi.ai/billing/pricing-and-success-packages | 20 November 2026 |
| 5 requests per turn; 150 output tokens a minute; 50% cache; two audio channels; long calls cost more | Vapi | docs.vapi.ai/assistants/model-intelligence/understanding-cost | 20 November 2026 |
| No on-premise deployments; server URL for function calling; bring your own models; volume pricing | Vapi | docs.vapi.ai/faq | 20 November 2026 |
| Deepgram Flux as built-in turn detection; 0.4-second default wait | Vapi | docs.vapi.ai/customization/speech-configuration | 20 November 2026 |
| Deepgram models on Vapi incl. flux-general-en and flux-general-multi; Aura and Aura-2 voices | Vapi | docs.vapi.ai/providers/transcriber/deepgram and docs.vapi.ai/providers/voice/deepgram | 20 November 2026 |
| Built-in tools: transfer, end call, SMS, keypad digits, voicemail, MCP | Vapi | docs.vapi.ai/tools | 20 November 2026 |
| US local $0.0085/min in, $1.15/month number; Media Streams $0.0044; AMD $0.0075/call | Twilio | twilio.com/en-us/voice/pricing/us | 20 November 2026 |
| Partial minutes rounded up on Programmable Voice and Media Streams | Twilio | help.twilio.com/articles/223132307 | 20 November 2026 |
| All route totals, per-minute figures, the $0.0195 headroom and the threshold table | Zian (our arithmetic from the rows above) | This page | 20 November 2026 |
Vapi and Deepgram change prices and models often; every figure here is as at 20 November 2026. If you would rather run answered calls and booked appointments than price a voice pipeline, Apply For Partnership.