Quick answer: Deepgram and ElevenLabs compete at three voice-agent layers. On list prices read 22 November 2026, ElevenLabs is cheaper for speech-to-text (Scribe v3 Live $0.22 an hour, Deepgram Flux STT $0.39), Deepgram for the voice (Aura-2 $0.030 per 1,000 characters, ElevenLabs $0.04) and for the bundled agent minute with your own LLM ($0.065 against $0.08). On a 10,000-minute month, mixed parts cost $186.67 before orchestration; the bundles with your own LLM cost $650.00 (Deepgram) to $799.96 (ElevenLabs Agents on Pro).
“Deepgram vs ElevenLabs” is really three questions, because the two companies now sell the same three things. Both sell streaming speech-to-text (STT): Deepgram Flux STT and Nova-3 against ElevenLabs Scribe v3 Live. Both sell text-to-speech (TTS) tuned for live calls: Deepgram Flux TTS and Aura-2 against ElevenLabs Eleven v4 Turbo, Flash v2.5 and v3 Conversational. And both sell a full agent API that wraps speech, a large language model (LLM) and turn-taking into one per-minute price: the Deepgram Voice Agent API against ElevenLabs Agents, which ElevenLabs also calls ElevenAgents. This page prices each layer from each vendor’s own pricing page and documentation, all read on 22 November 2026, in US dollars, then works one bill seven ways so you can see when buying one vendor’s bundle beats mixing the two. Telephony from Twilio, Plivo or a SIP trunk, the OpenAI, Anthropic or Google model you choose, and frameworks such as Pipecat and LiveKit sit outside both vendors’ speech prices, and we say where each one enters the bill.
Where Zian fits: an AI agents platform, not a speech API
If what you need is calls answered and appointments booked, rather than a speech pipeline to assemble, 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, and it is not a speech API you build on. ElevenLabs also sells a packaged receptionist of its own: its documentation describes Reception.ai as “an AI phone receptionist for small and medium businesses” that answers inbound calls and books appointments. This page compares the two vendors’ developer products, not that app.
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 works in 30+ languages, supports 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.
Deepgram vs ElevenLabs: the three places they overlap
The table below is the decision in one view. Each row is one layer, priced in the unit each vendor publishes, with what that unit actually measures. The units differ by layer, which is why a single “which is cheaper” answer does not exist.
| Layer | Deepgram product and published price | ElevenLabs product and published price | What the unit measures | Lower list price at this layer |
|---|---|---|---|---|
| Speech-to-text, streaming | Flux STT English $0.0065 a minute (shown as promo pricing, was $0.0077); Flux STT Multilingual $0.0078; Nova-3 Monolingual $0.0048 (was $0.0077); Nova-3 Multilingual $0.0058 (was $0.0092) | Scribe v3 Live and Scribe v2 Realtime $0.22 an hour (about $0.00367 a minute) | Deepgram: per second of audio, no block rounding. ElevenLabs: “billed per audio minute” | ElevenLabs, against every Deepgram streaming model |
| Text-to-speech for agents | Flux TTS $0.0450 per 1,000 characters; Aura-2 $0.030; Aura-1 $0.0150 | Eleven v4 Turbo, v3 Conversational and Flash / Turbo $0.04 per 1,000 characters; Eleven v4, v3 and Multilingual v2 $0.08 | Characters of input text, so the bill follows how much the agent says, not call length | Deepgram Aura-2 and Aura-1; ElevenLabs’ $0.04 models are below Flux TTS |
| Full agent API | Voice Agent API: Standard $0.075 a minute (STT, managed LLM and TTS); Custom, BYO LLM $0.065; Custom, BYO LLM and TTS $0.050; Advanced $0.163 | ElevenLabs Agents: plans from $0 (15 minutes) to $990 a month (12,375 minutes), then $0.080 a minute, LLM billed on top “at cost” | Deepgram: WebSocket connection time. ElevenLabs: call connection duration, with a 95% discount on silences longer than 10 seconds | Deepgram, at every tier up to Standard |
The quotable version: on list prices, ElevenLabs is cheaper at Deepgram’s home layer, speech-to-text, and Deepgram is cheaper at ElevenLabs’ home layer, the voice. All prices are pay-as-you-go, as at 22 November 2026, from deepgram.com/pricing, elevenlabs.io/pricing/api and elevenlabs.io/pricing/agents.
Speech-to-text: Deepgram Flux and Nova-3 vs ElevenLabs Scribe v3 Live
Deepgram sells two streaming families. Flux STT is its conversational model with “model-integrated end-of-turn detection”, offered as flux-general-en for “English (all accents)” and flux-general-multi for ten languages: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch. Deepgram’s Voice Agent API page says Flux “detects the end of the turn in about 260 ms”. Nova-3 is its general model without turn detection; its single-language list on the Models and languages overview carries 99 languages as at 22 November 2026 (100 lines, because English (Australia, New Zealand) now has its own line as well as sitting under English), and its multilingual mode covers 55 languages since 23 October 2026. Deepgram’s marketing pages describe this as “60+ languages”. The fuller per-language picture for a Deepgram agent is on our Deepgram voice agent languages and dialects matrix.
ElevenLabs’ current streaming model is Scribe v3 Live, released into the Speech to Text API on 9 November 2026 as the successor to Scribe v2 Realtime, which remains available. ElevenLabs lists it at 90+ languages, with turn prediction (the server commits a segment when the model predicts the speaker has finished, instead of waiting for silence alone), a multilingual mode for switches in the middle of a sentence, and a keyterm library of up to 10,000 terms. Its documentation claims a 44% lower semantic word error rate than Scribe v2 Realtime on the FLEURS benchmark; the pricing card for the same model says “43% fewer errors”. On 17 November 2026 ElevenLabs set every Scribe model at $0.20 an hour for batch and $0.22 for realtime on self-serve and Business plans, with Enterprise pricing unchanged.
Price per streamed minute, pay as you go: Scribe v3 Live about $0.00367, Nova-3 Monolingual $0.0048, Flux STT English $0.0065. Deepgram labels its speech-to-text table “Promo pricing” and shows the previous rates beside it; the page’s own data gives the promotion an end date of 13 February 2027. At the previous rates, Flux STT English and Nova-3 Monolingual are both $0.0077 a minute. Either way, Scribe v3 Live is the lower list price for the caller’s side of the call. Keyterms cost extra on both sides: ElevenLabs bills keyterm prompting and the Scribe v3 Live keyterm library at $0.08 an hour each on realtime models, and Deepgram bills Keyterm Prompting at $0.0013 a minute, so with one keyterm feature switched on Scribe v3 Live is $0.0050 a minute against $0.0061 for Nova-3 Monolingual.
Two limits cut the other way. ElevenLabs says “The official SDKs don’t support turn prediction, multilingual mode or the keyterm library yet”, so the features that make Scribe v3 Live an agent model need a direct WebSocket connection. And ElevenLabs “has not yet published a latency figure for Scribe v3 Live”; its ~150ms figure belongs to Scribe v2 Realtime and measures median time to a committed transcript. How each model decides a caller has stopped talking is compared on our STT turn detection page for Flux, Scribe and Speechmatics.
Text-to-speech: Deepgram Flux TTS and Aura-2 vs ElevenLabs v4 Turbo and Flash
Deepgram’s agent voice is now Flux TTS, which its Voice Agent API uses by default when no voice is set, in “English and 41 other languages”. Deepgram’s documentation marks the newer languages as Early Access without text normalisation, so it tells you to “prompt your LLM to write numbers, dates, and abbreviations out in words”, and its 23 October 2026 changelog says those 41 languages were not yet on the Australia endpoint. Aura-2 is the older family. Deepgram’s Voices and languages page lists Aura-2 voices in seven languages (English, Spanish, Dutch, French, German, Italian and Japanese) and Aura-1 in English only, and its agent documentation recommends Aura for “the broadest Deepgram voice catalog across English and Spanish”. Flux TTS costs $0.0450 per 1,000 characters, Aura-2 $0.030 and Aura-1 $0.0150, with a Flux TTS credit match of “$1 in credits for every $1 spent, up to $500, through Dec. 31, 2026”.
ElevenLabs’ agent default is Eleven v4 Turbo, which its API reference sets as the default agent voice model and its help centre confirms for new agents, at 90+ languages and $0.04 per 1,000 characters. Flash v2.5 (32 languages) and Eleven v3 Conversational (70+ languages) are also $0.04; Eleven v4, Eleven v3 and Multilingual v2 are $0.08. ElevenLabs’ own pages give a language count that depends on the model you pick: its help centre says an agent “can use every language supported by the text to speech model your agent uses”. An older ElevenLabs language guide still says a new agent starts on Flash v2 for English-only replies, so check which model your agent actually carries.
At 500 characters of agent speech per call minute, the voice costs $0.0075 a minute on Aura-1, $0.015 on Aura-2, $0.020 on any $0.04 ElevenLabs model and $0.0225 on Flux TTS. Latency claims do not settle the choice, because they measure different things: Deepgram says Flux TTS responses “start in as low as 80ms” and labels the figure time to first audio, while ElevenLabs gives v4 Turbo and Flash v2.5 a median of about 100ms that is server-side and excludes network and application time. ElevenLabs’ pricing page prints different figures from its own docs (Flash “~75ms” and v3 Conversational “~280ms”, against ~100ms and ~165ms in the docs), and ElevenLabs states its figures “are not comparable with time-to-first-byte (TTFB) or TTFA figures measured from a client”.
Deepgram Voice Agent API vs ElevenLabs Agents: what each bundle includes
Both bundles run listening, a model and speaking behind one connection, and both let you bring your own LLM. They differ in what else is inside and in whose speech models you may plug in.
| Question | Deepgram Voice Agent API | ElevenLabs Agents (ElevenAgents) |
|---|---|---|
| Price per minute | $0.050 to $0.163 by tier; Standard $0.075 includes a managed Standard-tier model | $0.080 beyond the plan’s included minutes; the LLM is extra, passed through “with no markup”, except that select Gemini and Claude models are billed at Vertex AI regional rates, “a 10% increase” |
| Speech-to-text inside the agent | Deepgram only: Flux STT or Nova (“Currently only Deepgram is supported”) | ElevenLabs only in the API’s list: elevenlabs, scribe_realtime (default), scribe_v2_turbo, scribe_v3_live, scribe_v3_live_medical |
| Voice inside the agent | Flux TTS (default), Aura, Deepgram-managed Cartesia, or your own OpenAI, ElevenLabs, Cartesia or Amazon Polly account | ElevenLabs models: Turbo v2, Turbo v2.5, Flash v2, Flash v2.5, Multilingual v2, v3 Conversational, v4, v4 Turbo (default) |
| Can it use the other vendor’s speech? | Yes for the voice: “We support any of ElevenLabs’ Turbo 2.5 voices”, and eleven_v3 over HTTPS | Not among the listed speech-to-text options |
| Phone calls | Guides for Twilio, Plivo, Genesys Cloud CX, Amazon Connect and AudioCodes; for Twilio and Plivo your server bridges carrier audio to the WebSocket, AudioCodes runs through a Deepgram open-source bridge that you host, and Genesys Cloud CX connects to a Deepgram-hosted Audio Connector endpoint | Native Twilio integration (purchased numbers inbound and outbound, verified caller IDs outbound only), SIP trunking, and setup guides for Vonage, Telnyx, Plivo, Exotel, Bandwidth, Sinch, Vobiz, Voximplant, Amazon Connect, Genesys, Five9 and Microsoft Teams |
| Concurrent calls | 45 on Pay as you go in every region; on Growth up to 60 in North America and 45 in Europe, Australia and India | 4, 6, 10, 20, 30 or 40 by plan; burst to three times your limit at double the rate |
| How the stages run | Deepgram “transcribes it, passes the text to the language model, converts the model’s reply back to speech”, behind one connection | Custom mode chains the same three stages; Eleven Realtime v1 Mini and v1 Max are speech-to-speech models at $0.08 a minute plus token charges, in every workspace since 2 November 2026 |
The asymmetry in the fourth row matters for a mixed stack. Deepgram’s agent will speak with an ElevenLabs voice on your own ElevenLabs account, on the cheaper “BYO TTS” tiers, but ElevenLabs Agents lists only ElevenLabs speech recognition. If you want Scribe listening and an Aura-2 voice, neither bundle does it; you assemble that pipeline yourself. How the Deepgram bridge server and Twilio charges add up is itemised on our Vapi vs Deepgram Voice Agent API cost breakdown, and why ElevenLabs’ higher plans buy concurrency rather than cheaper minutes is on our Vapi vs ElevenLabs Agents comparison.
Apply For Partnership if you would rather have inbound and outbound agents that answer, book and follow up across calls, SMS, email, WhatsApp, Instagram DMs and Facebook Messenger than wire speech vendors together.
What does a 10,000-minute month cost on Deepgram, ElevenLabs or both?
One workload, priced seven ways. The inputs: 5,000 calls a month, each exactly 2 minutes, so 10,000 call minutes and no rounding question. Speech-to-text listens to the caller for the whole call, so 10,000 STT minutes. The agent speaks 500 characters per call minute, so 5,000,000 characters; ElevenLabs’ own calculator treats 1,000 characters as about a minute of audio, so this is an agent talking roughly half the call. Your LLM, your carrier and, for the do-it-yourself rows, the server that runs turn-taking and barge-in are the same or unpriced in every row, so they are left out. Pay-as-you-go list prices, monthly billing, US dollars, as at 22 November 2026.
| Stack | Speech-to-text | Voice | Agent fee | Monthly total | Per call minute |
|---|---|---|---|---|---|
| A. Mixed parts: Scribe v3 Live listens, Aura-2 speaks | 166.67 hours × $0.22 = $36.67 | 5,000 × $0.030 = $150.00 | None (you orchestrate) | $186.67 | $0.0187 |
| B. ElevenLabs parts: Scribe v3 Live and Eleven v4 Turbo | $36.67 | 5,000 × $0.04 = $200.00 | None (you orchestrate) | $236.67 | $0.0237 |
| C. Deepgram parts: Flux STT and Flux TTS | 10,000 × $0.0065 = $65.00 | 5,000 × $0.045 = $225.00 | None (you orchestrate) | $290.00 | $0.0290 |
| D. Mixed parts: Flux STT listens, Eleven v4 Turbo speaks | $65.00 | $200.00 | None (you orchestrate) | $265.00 | $0.0265 |
| E. Deepgram Voice Agent API, Custom BYO LLM | Included | Included (Deepgram voice) | 10,000 × $0.065 = $650.00 | $650.00 | $0.0650 |
| F. Deepgram Voice Agent API, BYO LLM and TTS, with an ElevenLabs Turbo v2.5 voice | Included | $200.00, paid to ElevenLabs | 10,000 × $0.050 = $500.00 | $700.00 | $0.0700 |
| G. ElevenLabs Agents, Pro plan, your own LLM | Included | Included | $99 + 8,762 × $0.08 = $799.96 | $799.96 | $0.0800 |
Three checks on the table. First, ElevenLabs Agents lands at almost the same total on Creator ($22 + 9,725 × $0.08 = $800.00) and Scale ($299 + 6,262 × $0.08 = $799.96) as on Pro; the plan choice changes your concurrent-call ceiling (10, 20 or 30), not the bill, and Business is $990.00 flat at this volume. Second, at Deepgram’s pre-promotion speech-to-text rate of $0.0077, row C rises to $302.00 and row D to $277.00; rows E and F do not change. Third, if you let each vendor supply the model, Deepgram Standard at $0.075 is $750.00 all in, which is less than ElevenLabs Agents’ $799.96 before ElevenLabs adds any LLM charge.
The Orchestration Premium Rule: when one vendor’s bundle beats mixing both
The gap between a vendor’s bundle and the same vendor’s own parts is the price of the work the bundle does for you: turn-taking, barge-in, session state, function-call plumbing and hosting. On our workload that premium is $360.00 a month for Deepgram (row E $650.00 against row C $290.00, or $0.036 a minute) and $563.29 a month for ElevenLabs (row G $799.96 against row B $236.67, or about $0.056 a minute).
The Orchestration Premium Rule: buy one vendor’s agent API when running your own turn-taking, barge-in and hosting would cost you more than $0.036 a minute (Deepgram) or $0.056 a minute (ElevenLabs); below that, buy parts, and once you are buying parts, buy each layer from whichever vendor prices it lower.
The parts never catch the bundle on list price alone. Deepgram’s bundle only matches its own Flux STT and Flux TTS if the agent speaks 1,300 characters per call minute, and ElevenLabs’ bundle only matches Scribe v3 Live plus v4 Turbo at about 1,908 characters per call minute; both are more speech than ElevenLabs’ own estimate of a full minute of audio. So the question is never “are the parts cheaper”, which they are, but “is my engineering time cheaper than the premium”. At 10,000 minutes the premium is a few hundred dollars a month; at 100,000 minutes on the same rates it is roughly ten times that, and that larger figure is what to weigh your own engineering time against before you run a framework such as Pipecat or LiveKit around separately bought speech.
Which should you pick for each layer?
How we chose: a vendor counts as offering a layer when its own pricing page lists a per-unit price for it and its own documentation describes using it in a live voice agent; Deepgram and ElevenLabs both meet that bar at all three layers. Prices are pay-as-you-go list prices read on 22 November 2026, compared in each layer’s own unit. Disclosure: zian.ai publishes this comparison and lists its own platform in it, first, as the option for teams that would rather not build.
| If you are… | Pick | Why, from the figures above | Wrong for you if… |
|---|---|---|---|
| A business that wants calls answered and appointments booked, not a speech 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; if you want a packaged inbound receptionist from one of these two vendors instead, ElevenLabs sells Reception.ai |
| A developer team that wants one connection and the lowest bundled minute | Deepgram Voice Agent API | $0.065 with your LLM, or $0.075 with a managed Standard-tier model; 45 connections on Pay as you go | You cannot run a bridge server between your carrier and the WebSocket (Genesys Cloud CX, which connects to a Deepgram-hosted endpoint, is the exception) |
| A team that wants native Twilio and SIP numbers and ElevenLabs voices in one product | ElevenLabs Agents | Twilio and SIP connect to the agent directly; v4 Turbo voice and Scribe recognition are inside the $0.08 minute | You need more than 40 simultaneous calls without Enterprise or burst pricing |
| A team building its own pipeline that wants the lowest speech-to-text list price | ElevenLabs Scribe v3 Live | $0.22 an hour, about $0.00367 a minute, against $0.0048 to $0.0078 for Deepgram’s streaming models | You rely on an official SDK for turn prediction, or need a published latency figure for the exact model |
| A team that wants Deepgram’s turn detection in its own pipeline | Deepgram Flux STT | Model-integrated end of turn, about 260 ms per Deepgram, $0.0065 a minute on current promo pricing | Your callers speak a language outside Flux’s ten |
| A team building its own pipeline that wants the lowest voice list price | Deepgram Aura-1 or Aura-2 | $0.0150 and $0.030 per 1,000 characters, against $0.04 for ElevenLabs’ agent models and $0.045 for Flux TTS | You need Aura-1 outside English, or Aura-2 outside its seven languages: English, Spanish, Dutch, French, German, Italian and Japanese |
| A team whose callers speak many languages and wants one real-time voice model | ElevenLabs Eleven v4 Turbo | 90+ languages at $0.04 per 1,000 characters, against Flux TTS’s English plus 41 others | Your budget is set on Aura-2’s $0.030 |
The sentence on each vendor’s pages that cuts against this page
Against the mixed-parts result: Deepgram’s Voice Agent API page says its speech-to-text, text-to-speech and orchestration “all run through one API, so you don’t have to build and maintain the voice stack yourself”. Rows A to D exclude exactly that work, which is why we price it as a premium rather than calling the parts cheaper outright.
Against Scribe v3 Live as the speech-to-text pick: ElevenLabs says its official SDKs do not yet support turn prediction or multilingual mode, and it has not published a latency figure for the model.
Against Aura-2 as the voice pick: Deepgram itself makes Flux TTS, not Aura-2, the Voice Agent API default, and Flux TTS costs half as much again per character.
Against Deepgram’s speech-to-text prices: they are labelled promo pricing, and the page data dates the promotion’s end at 13 February 2027.
Against ElevenLabs’ per-request pricing: its 26 October 2026 changelog says “Text to Speech and Speech to Text requests priced below one credit are charged one credit”, while its API pricing FAQ says “API usage is billed in US dollars, not credits”. Check your own usage dashboard before you model many very short utterances.
What this comparison leaves out
We do not score accuracy or voice quality. Each vendor publishes its own benchmarks against rivals, and we have not tested them on your calls. We leave out the LLM, which costs the same in rows A to G when you bring your own, and telephony, which depends on your carrier and on whether you connect through Twilio Media Streams, a SIP trunk or ElevenLabs’ native Twilio integration. Enterprise, Growth and yearly plan discounts, ElevenLabs’ silence discount and Deepgram’s Flux TTS credit match are also left out; each would lower one or more rows. Re-price on the day you sign, because both vendors changed prices or models in the four weeks before this page was written.
Frequently asked questions
Is Deepgram or ElevenLabs cheaper for a voice agent?
It depends on the layer. On list prices read 22 November 2026, ElevenLabs’ Scribe v3 Live is cheaper for speech-to-text at $0.22 an hour, Deepgram’s Aura-2 is cheaper for the voice at $0.030 per 1,000 characters, and Deepgram’s Voice Agent API costs less per minute than ElevenLabs Agents at its Standard and Custom tiers, as Deepgram’s pricing page and ElevenLabs’ agents pricing page show. On our 10,000-minute month the bundles cost $650.00 and $799.96.
Can I use an ElevenLabs voice inside Deepgram’s Voice Agent API?
Yes. Deepgram’s documentation says it supports “any of ElevenLabs’ Turbo 2.5 voices”, on your own ElevenLabs account, and eleven_v3 over HTTPS. You pay Deepgram $0.056 a minute on Standard with your own voice, or $0.050 with your own LLM and voice, plus ElevenLabs’ character charge.
Can ElevenLabs Agents use Deepgram speech-to-text?
Not as a listed option. ElevenLabs’ Create agent API reference lists elevenlabs, scribe_realtime, scribe_v2_turbo, scribe_v3_live and scribe_v3_live_medical as speech-to-text providers. To pair Deepgram listening with an ElevenLabs voice, build the pipeline yourself or use Deepgram’s agent with an ElevenLabs voice.
Which is faster, Deepgram or ElevenLabs?
Their published figures measure different things. Deepgram says Flux TTS starts audio “in as low as 80ms” and Flux ends a turn in about 260 ms; ElevenLabs gives v4 Turbo and Flash v2.5 a median of about 100ms, server-side, excluding network time. ElevenLabs says its figures are not comparable with client-measured ones, so measure both from your own region.
Which supports more languages, Deepgram or ElevenLabs?
Each publishes its own list. Deepgram Nova-3 lists 99 languages, Flux STT Multilingual 10, Flux TTS English plus 41 others and Aura-2 seven. ElevenLabs lists Scribe v3 Live and Eleven v4 Turbo at 90+ languages each and Flash v2.5 at 32.
Are Deepgram’s speech-to-text prices permanent?
No. Deepgram labels them promo pricing and shows the earlier rates beside them: Flux STT English $0.0065 a minute, was $0.0077, and Nova-3 Monolingual $0.0048, was $0.0077. The pricing page’s data gives the promotion an end date of 13 February 2027.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Flux STT English $0.0065 (was $0.0077), Flux STT Multilingual $0.0078, Nova-3 Monolingual $0.0048 (was $0.0077), Nova-3 Multilingual $0.0058 (was $0.0092); “Promo pricing”; promotion end 13 February 2027 (page data); per-second billing, no block rounding; Keyterm Prompting add-on $0.0013 a minute | Deepgram | deepgram.com/pricing | 22 November 2026 |
| Flux TTS $0.0450, Aura-2 $0.030, Aura-1 $0.0150 per 1,000 characters; Flux TTS credit match to 31 December 2026 | Deepgram | deepgram.com/pricing | 22 November 2026 |
| Voice Agent tiers $0.075 / $0.056 / $0.065 / $0.050 / $0.163 / $0.122; tier descriptions; WebSocket connection time; Growth concurrency 60 North America, 45 Europe, Australia and India | Deepgram | deepgram.com/pricing | 22 November 2026 |
| Flux end of turn “about 260 ms”; “all run through one API”; Global, EU, India or Australia endpoints | Deepgram | deepgram.com/product/voice-agents | 22 November 2026 |
| Flux TTS “as low as 80ms” time to first audio; “60+ languages” (marketing) | Deepgram | deepgram.com/product/text-to-speech; deepgram.com/product/speech-to-text | 22 November 2026 |
| flux-general-en “English (all accents)”; flux-general-multi 10 languages; Nova-3 single-language list (99 languages, 100 lines) | Deepgram | developers.deepgram.com/docs/models-languages-overview | 22 November 2026 |
| Agent speech-to-text is Deepgram only (Flux STT or Nova); “Currently only Deepgram is supported” | Deepgram | developers.deepgram.com/docs/voice-agent-stt-models; /docs/configure-voice-agent | 22 November 2026 |
| Flux TTS default, English and 41 other languages; Early Access without text normalisation; Aura for “the broadest Deepgram voice catalog across English and Spanish”; managed Cartesia; BYO OpenAI, ElevenLabs, Cartesia, Amazon Polly; “any of ElevenLabs’ Turbo 2.5 voices”; eleven_v3 over HTTPS | Deepgram | developers.deepgram.com/docs/voice-agent-tts-models | 22 November 2026 |
| Aura-2 voices in English, Spanish, Dutch, French, German, Italian and Japanese; Aura-1 English only | Deepgram | developers.deepgram.com/docs/tts-models | 22 November 2026 |
| Standard-tier and Advanced-tier managed models; BYO endpoints | Deepgram | developers.deepgram.com/docs/voice-agent-llm-models | 22 November 2026 |
| 45 Voice Agent connections on Pay as you go in every region | Deepgram | developers.deepgram.com/reference/api-rate-limits | 22 November 2026 |
| Twilio bridge via bidirectional media stream; telephony guides for Twilio, Plivo, Genesys, Amazon Connect, AudioCodes; Genesys Audio Connector URI hosted by Deepgram; open-source AudioCodes bridge | Deepgram | developers.deepgram.com/docs/twilio-and-deepgram-voice-agent; /docs/voice-agent-architecture; /docs/genesys-and-deepgram-voice-agent; /docs/integrate-deepgram-voice-agent-with-audiocodes | 22 November 2026 |
| Nova-3 multilingual 55 languages (23 October 2026); Flux TTS 41 languages not yet on Australia endpoint (23 October 2026); 36 Nova-3 languages (16 November 2026) | Deepgram | developers.deepgram.com/changelog | 22 November 2026 |
| Scribe v3 Live and Scribe v2 Realtime $0.22 an hour; Scribe v3 $0.20; v4 Turbo, v3 Conversational, Flash / Turbo $0.04; v4, v3, Multilingual v2 $0.08 per 1,000 characters; “billed per audio minute”; “billed in US dollars, not credits”; realtime keyterm prompting and Scribe v3 Live keyterm library $0.08 an hour each; ~1,000 characters ≈ 1 minute in the calculator; card figures Flash ~75ms, v3 Conversational ~280ms, Scribe v3 Live “43% fewer errors” | ElevenLabs | elevenlabs.io/pricing/api | 22 November 2026 |
| Plans $0 / $6 / $22 / $99 / $299 / $990; 15 to 12,375 included minutes; 4 to 40 concurrent calls; $0.080 extra minute; $0.160 burst, up to three times the concurrency limit at double the rate; LLM and telephony “At cost” | ElevenLabs | elevenlabs.io/pricing/agents | 22 November 2026 |
| Connection-duration billing; 95% discount for silence over 10 seconds | ElevenLabs | elevenlabs.io/docs/help-center/product/eleven-agents/how-much-does-eleven-agents-cost | 22 November 2026 |
| Scribe v3 Live 90+ languages, turn prediction, multilingual mode, 10,000-term keyterm library, 44% lower semantic WER; SDK limitation; v4 Turbo 90+ languages ~100ms; Flash v2.5 32 languages ~100ms; v3 Conversational 70+ languages ~165ms; “Median, excluding application & network latency” | ElevenLabs | elevenlabs.io/docs/models | 22 November 2026 |
| Server-side medians, not comparable with client TTFB or TTFA; no published Scribe v3 Live figure; Scribe v2 Realtime ~150ms to committed transcript | ElevenLabs | elevenlabs.io/docs/eleven-api/concepts/latency-methodology | 22 November 2026 |
| Agent speech-to-text providers list; agent voice model list, default eleven_v4_turbo | ElevenLabs | elevenlabs.io/docs/eleven-agents/api-reference/agents/create | 22 November 2026 |
| Agent languages follow the voice model; new agents use Eleven v4 Turbo | ElevenLabs | elevenlabs.io/docs/help-center/product/eleven-agents/which-languages-can-i-use-with-eleven-agents | 22 November 2026 |
| LLM pass-through “with no markup”; custom LLM support | ElevenLabs | elevenlabs.io/docs/eleven-agents/customization/llm | 22 November 2026 |
| Twilio purchased numbers inbound and outbound; verified caller IDs outbound only | ElevenLabs | elevenlabs.io/docs/eleven-agents/phone-numbers/twilio-integration/native-integration | 22 November 2026 |
| Telephony and contact-centre guides: Vonage, Telnyx, Plivo, Exotel, Bandwidth, Sinch, Vobiz, Voximplant, Amazon Connect, Genesys, Five9, Microsoft Teams | ElevenLabs | elevenlabs.io/docs/eleven-agents (Phone numbers section) | 22 November 2026 |
| Reception.ai, “an AI phone receptionist for small and medium businesses” | ElevenLabs | elevenlabs.io/docs/reception-ai/overview | 22 November 2026 |
| Eleven Realtime v1 Mini and v1 Max, $0.08 a minute plus tokens, all workspaces (2 November 2026); Scribe v3 and v3 Live released (9 November 2026); Scribe $0.20 / $0.22 (17 November 2026); one-credit minimum (26 October 2026) | ElevenLabs | elevenlabs.io/docs/changelog | 22 November 2026 |
| Rows A to G totals, per-minute figures, $360.00 and $563.29 premiums, 1,300 and 1,908 character crossovers, $302.00 and $277.00 pre-promotion rows, $750.00 Deepgram Standard | Zian (our arithmetic from the rows above) | This page | 22 November 2026 |
Apply For Partnership to see whether Zian’s inbound and outbound agents fit your call volume, without choosing a speech vendor first.