We read the pricing pages, docs and license files of 16 voice agent products on October 19, 2026 and October 20, 2026: 7 hosted voice agent platforms, 5 model-provider realtime APIs, 2 open source frameworks and 2 cloud contact-center services. Every value carries a verbatim quote and its source. Prices stay in each vendor's own unit, so they are grouped by unit and never ranked across units.
Hume EVI: shutting down November 13, 2026.
Ultravox Realtime: acquired by Inworld.
OpenAI Realtime API: beta interface: move to GA.
Synthflow: enterprise pricing only.
Amazon Connect Customer and Twilio ConversationRelay list a per-minute rate without saying what it covers.
Hume's EVI overview says Hume is sunsetting its TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, both APIs remain fully supported until then, and any data in the account is permanently deleted after November 13, 2026.
“Hume is sunsetting the TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, and both APIs remain fully supported until then. Any data in your account is permanently deleted after November 13, 2026.”
A shutdown with a date, an acquisition, an interface migration and a pricing page that now lists one enterprise contract, each said only as far as the vendor's own page says it, with the quote and the link. Where the page gives no date, neither do we.
Hume's EVI overview says Hume is sunsetting its TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, both APIs remain fully supported until then, and any data in the account is permanently deleted after November 13, 2026.
“Hume is sunsetting the TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, and both APIs remain fully supported until then. Any data in your account is permanently deleted after November 13, 2026.”
Ultravox's pricing page says Ultravox has been acquired by Inworld to accelerate its speech-to-speech research and voice agent capabilities. No date given.
“Ultravox has been acquired by Inworld to accelerate its speech-to-speech research and voice agent capabilities.”
OpenAI's Realtime guide tells anyone with a beta Realtime integration to migrate it to the GA interface before moving forward with new work. No date given.
“If you still have a beta Realtime integration, migrate it to the GA interface before moving forward with new work.”
Synthflow's pricing page says Enterprise contracts start at $30,000 annually, scoped around call volume, concurrency, telephony setup, integrations, security needs and launch support. The page shows no free tier or self-serve plan. No date given.
“Enterprise contracts start at $30,000 annually. Final pricing is scoped around call volume, concurrency, telephony setup, integrations, security needs, and launch support.”
Each card shows how the product charges in its own words and what that price includes, then phone numbers, SIP, model choice, function calling, self-hosting and license. Every value links the page it came from.
Hosted voice agent platforms
7 products
Managed platforms that run the whole call: turn-taking, phone numbers or SIP, and the speech and language models, billed per minute, per plan or per contract.
Enterprise contracts from $30,000 annually- per year (contract minimum)
Includes: Scoped around call volume, concurrency, telephony setup, integrations, security needs and launch support (stated)
“Enterprise contracts start at $30,000 annually. Final pricing is scoped around call volume, concurrency, telephony setup, integrations, security needs, and launch support.”
Speech-to-speech or voice agent APIs from a model provider, billed per audio token, per minute of session or per minute of connection. Telephony is usually yours to bring.
Standard: $0.075/min (Pay As You Go), $0.068/min (Growth)- per minute of websocket connection time
Includes: See the Voice Agent API billing note
“Tier · Pay As You Go · Growth; Standard · $0.075/min i calculated based on websocket connection time. · $0.068/min i calculated based on websocket connection time.”
Includes: Not stated. The pricing row does not say what the $0.07 per minute covers.
“Conversation Relay Integrate voice AI into your tech stack without the complexity. Focus on your AI while we handle the voice infrastructure. · $0.07/ min”
Every price line, in the vendor's own words, grouped by what it charges for, with what the vendor says it includes. Rows are sorted by amount only inside one unit; a call minute, an agent minute, an audio token, a plan and a platform fee are different things, so nothing is converted or ranked across units.
Per minute
Per-minute rates in each vendor's own words: a whole call, talk time, a voice minute, websocket connection time, a transfer minute or a single component such as hosting, voice infrastructure or telephony. What a minute covers differs by vendor, so read the Includes line.
Per minute: 24 price lines in 7 units, each in the vendor's own words, sorted only inside one unit
Third-party SIP minutes: then $0.004/min (Ship), $0.003/min (Scale)inferred
Inbound and outbound minutes using a third-party SIP trunk (row description)
“Third-party SIP minutes · 1,000 minutes included · 5,000 minutes included (then $0.004 per min) · 50,000 minutes included (then $0.003 per min) · Custom”
Not stated. The pricing row does not say what the $0.07 per minute covers.
“Conversation Relay Integrate voice AI into your tech stack without the complexity. Focus on your AI while we handle the voice infrastructure. · $0.07/ min”
Standard: $0.075/min (Pay As You Go), $0.068/min (Growth)
See the Voice Agent API billing note
“Tier · Pay As You Go · Growth; Standard · $0.075/min i calculated based on websocket connection time. · $0.068/min i calculated based on websocket connection time.”
Hosted framework clouds that bill the time an agent runs (LiveKit agent session minutes, Pipecat Cloud agent minutes), separate from telephony and models.
Per agent or agent session minute: 3 price lines in 2 units, each in the vendor's own words, sorted only inside one unit
Scoped around call volume, concurrency, telephony setup, integrations, security needs and launch support (stated)
“Enterprise contracts start at $30,000 annually. Final pricing is scoped around call volume, concurrency, telephony setup, integrations, security needs, and launch support.”
LLM, STT and TTS from 80+ providers, billed to the provider (stated)
“Connect directly with 80+ service providers for LLM, STT, and TTS including Anthropic, Cartesia, Deepgram, ElevenLabs, Google, OpenAI, and more. See full list here. Billed to provider”
Text token pricing applies to transcription, tool calls, knowledge grounding and conversation history (stated)
“The text tokens input and output pricing applies to specific use cases such as speech-to-text transcription, tool calls for task completion or knowledge grounding, adding conversation history to the session, and so on.”
Billed on conversation duration; includes STT, LLM processing and TTS orchestration (stated)
“The Voice Agent API is billed based on the duration of the conversation. It includes the orchestration of Speech-to-Text (listening), LLM processing (thinking), and Text-to-Speech (speaking).”
The same ElevenAgents overview lists 5k+ voices across 31 languages in its voice and language row, and describes its TTS model as covering 5k+ voices and 70+ languages in the architecture section. Both are shown as stated and not reconciled.
Voice and language row
31 languages (voice & language guide row); architecture section says 70+ languages for TTS
“Select from 5k+ voices across 31 languages with customization options”
Every product against every field, each cell linking its source. Not stated is not the same as no. This is not a ranking.
Telephony, models and actions
Phone numbers, SIP and bring-your-own carrier support, whether you choose the models, tool calling, and transfers or agent handoffs, as each vendor states them.
Telephony, models and actions: 16 products by Phone numbers, SIP, BYO carrier, BYO models, Function calling, Transfer or handoff
Each product's status as its own page puts it, and the day its pages were read. Some vendors were read across two days: pricing on one, docs on the next.
Status for 16 products, as each page states it, with the day it was verified
“Amazon Nova Sonic is a speech understanding and generation model that delivers natural, human-like conversational AI with low latency and industry-leading price performance.”
“Deepgram's Voice Agent API handles the full speech pipeline (listening, thinking, and speaking) so you can focus on what your agent does, not how it hears or talks.”
Values whose meaning is narrower than they look: per-minute rates that do not say what they cover, an annual-rate note that does not say which prices it applies to, an estimator example, a promotion with no end date and a per-connection limit.
Amazon Connect is a full contact-center suite, not a voice-agent API. Its pricing page says standard telephony rates apply on top of the per-minute voice rate, and does not say which AI features that rate covers.
“Voice · $0.038 per voice min (standard telephony rates apply)”
The voice pricing row does not say what the per-minute ConversationRelay rate covers.
“Conversation Relay Integrate voice AI into your tech stack without the complexity. Focus on your AI while we handle the voice infrastructure. · $0.07/ min”
Ultravox's pricing page carries a note marking its prices as an annual rate, beside per-minute and per-month figures. It does not say which prices the note covers, so read the page before budgeting.
“Pro Good bye hard concurrency caps. Everything you need to start scaling your Voice AI business. $100 /month”
Values whose plan, tier or column name comes from the page layout rather than the quote, and one feature backed only by a docs navigation label, each tagged inferred wherever it renders.
VapiPricingPro: 10% of Vapi hosting fee, $999/mo minimumPlan name from page layout: the quote is the price block of the Pro card and does not carry the name.
VapiConcurrency4 concurrent calls (no success package)Plan from page layout: the quote pairs 1 included phone number with 4 concurrent calls and names no plan; the Core and Pro cards list more numbers.
VapiConcurrency10 concurrent calls (Core)Plan name from page layout: the quote sits on the Core card and does not name the plan.
VapiConcurrency30 concurrent calls (Pro)Plan name from page layout: the quote sits on the Pro card and does not name the plan.
Retell AIPricingEnterprise: Custom PricingPlan name from page layout: the quote is the custom-pricing card and does not carry the word Enterprise.
Bland AIPricingTransfer time: $0.05/min (Start), $0.04/min (Build)Plans assigned by table column order: the row lists $0.05/min, $0.04/min and Custom under columns read as Start, Build and Enterprise.
Bland AIConcurrency10 (Start), 50 (Build), Custom (Enterprise)Plans assigned by table column order: 10, 50 and Custom sit under columns read as Start, Build and Enterprise.
ElevenLabs AgentsPricingAdditional call minute $0.080; burst $0.160 per minute; text message $0.003Rates read from the plan table: the quote is the header row followed by the first plan's row, and the same three rates repeat on every plan row.
ElevenLabs AgentsFree tier or credits15 minutes of calls and 4 concurrent calls included (Free plan)Plan from page layout: these lines sit on the Free plan card; the quote does not name the plan.
ElevenLabs AgentsConcurrency4 (Free), 6 (Starter), 10 (Creator), 20 (Pro), 30 (Scale), 40 (Business) concurrent callsPlans assigned by table row order (Free, Starter, Creator, Pro, Scale, Business); the rows do not name them.
Cartesia Managed AgentsPricingCall duration: $0.06 per minute (Free through Scale plans)Plans assigned by table column order (Free through Scale, then Enterprise); the row does not name them.
Cartesia Managed AgentsFree tier or credits$1 prepaid agents per month (Free plan)Plan from page layout: the line sits on the Free plan card; the quote does not name the plan.
Cartesia Managed AgentsPhone numbersCartesia-provisioned phone numbers: 1, 3, 5, 10, Custom by planPlans assigned by table column order (Free through Scale, then Enterprise); the row does not name them.
Cartesia Managed AgentsConcurrencyConcurrent calls: 8, 12, 20, 60, Custom by planPlans assigned by table column order (Free through Scale, then Enterprise); the row does not name them.
Ultravox RealtimePricingSIP: .5¢ per minute (Pay As You Go), .48¢ per minute (Pro)Plans assigned by table column order (Pay As You Go, Pro, Enterprise); the row does not name them.
Ultravox RealtimeConcurrencyUp to 5 concurrent calls (Pay As You Go)Plan from page layout: the line sits on the Pay As You Go card; the quote does not name the plan.
Ultravox RealtimeConcurrencyNo hard caps on concurrency (Pro)Plan from page layout: the line sits on the Pro card; the quote does not name the plan.
OpenAI Realtime APISIPTelephony and SIP guide listed in the Realtime docs navigationBacked only by a navigation label in the Realtime docs ("Telephony and SIP"); the guide text does not itself describe SIP support.
Gemini Live APIPricingAudio input: $3.00 per 1M tokens or $0.005/min (paid tier)Tier from table layout: the price sits in the paid-tier column beside the free tier's Free of charge; the row does not name the tier.
Gemini Live APIPricingAudio output: $12.00 per 1M tokens or $0.018/min (paid tier), including thinking tokensTier from table layout: the price sits in the paid-tier column beside the free tier's Free of charge; the row does not name the tier.
Gemini Live APIFree tier or creditsInput price: Free of charge (free tier)Tier from table layout: Free of charge sits in the free-tier column; the row does not name the tier.
Deepgram Voice Agent APIPricingStandard - BYO TTS: $0.065/min (PAYG), $0.051/min (Growth)Columns from table layout: the prices sit under the Pay As You Go and Growth columns of the Voice Agent table; the row does not repeat the column names.
Deepgram Voice Agent APIPricingCustom - BYO LLM: $0.059/min (PAYG)Column from table layout: the price sits under the Pay As You Go column of the Voice Agent table; the row does not repeat the column name.
Deepgram Voice Agent APIPricingCustom - BYO LLM + TTS: $0.050/min (PAYG), $0.041/min (Growth)Columns from table layout: the prices sit under the Pay As You Go and Growth columns of the Voice Agent table; the row does not repeat the column names.
Deepgram Voice Agent APIPricingAdvanced: $0.163/min (PAYG), $0.146/min (Growth)Columns from table layout: the prices sit under the Pay As You Go and Growth columns of the Voice Agent table; the row does not repeat the column names.
LiveKit Agents + LiveKit CloudPricingAgent session minutes: 1,000 included (Build); 5,000 then $0.01/min (Ship); 50,000 then $0.01/min (Scale)Plans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudPricingThird-party SIP minutes: then $0.004/min (Ship), $0.003/min (Scale)Plans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudFree tier or credits$2.50 in LiveKit Inference credits on the Build planPlan from table column order: the credit row's first column is read as Build; the row does not name it.
LiveKit Agents + LiveKit CloudPhone numbersUS local numbers: 1 free, then $1.00/month (Ship, Scale)Plans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudRecording and transcriptsAgent session recordings: 1,000 min included (Build), then $0.005/min (Ship, Scale)Plans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudComplianceSOC 2 Type II and network pentest reports on Scale and EnterprisePlans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudComplianceHIPAA compliance on Scale and EnterprisePlans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
LiveKit Agents + LiveKit CloudConcurrencyConcurrent agent sessions: 5 (Build), 20 (Ship), up to 600 (Scale)Plans assigned by table column order (Build, Ship, Scale, Enterprise); the row does not name them.
Pipecat + Pipecat CloudSelf-hostingAgents run in a Kubernetes cluster you operate (Pipecat Enterprise)Column from page layout: the quote is a comparison row whose second column is read as Pipecat Enterprise; the row does not name it.
Not included
A product considered and left out, and a product listed without a price, each with the reason.
Speechmatics Flownot included
Speechmatics' own pages describe no current Flow voice-agent product: its Flow documentation now leads to Agent STT (a speech-to-text product for voice agents, not a full agent runtime), and its pricing page lists Agent STT (Linden 1) but no Flow price. Agent STT belongs in the speech-to-text directory. Speechmatics does not state that Flow is retired, so no status change is recorded.
Hume EVI pricinglisted without a price
Hume EVI is included, but no EVI price is stated: hume.ai/pricing does not list EVI rates on 2026-10-20, and the EVI overview gives none.
Guides on voice agents
Our longer reads on building voice agents and the APIs in this directory. Point-in-time posts; where a post and this directory differ, the directory is the newer read.
Read from each vendor's pricing pages, docs and license files on October 19, 2026 and October 20, 2026, with every quote string-matched against the page text that day. Each product shows the day its pages were read. Prices, plans and support status change; check the linked source before you commit.
Not a benchmark
Nothing here measures latency, interruption handling or call quality. Latency figures appear only under a Vendor-stated label. Shortlist on these facts, then place real calls.
Prices in the vendor's own unit
A call minute, a talk-time minute, a websocket minute, an agent session minute, an audio token, a plan and a platform fee are different things, and what a minute covers differs by vendor. Prices are grouped by unit and sorted only inside one unit, never converted and never compared across units.
Includes, as stated
Each price line says what the vendor says it covers, or Not stated where the page does not say. Amazon Connect and Twilio ConversationRelay list a per-minute rate without saying what it covers.
Not stated and inferred
Not stated means the pages we read did not say; it is not evidence the feature is missing. Inferred marks our reading, such as a plan name taken from a pricing table's column order, and always carries the reason.
FAQ
What changed in voice AI agent platforms recently?
In the vendors' own words: Hume EVI: "Hume is sunsetting the TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, and both APIs remain fully supported until then. Any data in your account is permanently deleted after November 13, 2026." Ultravox Realtime: "Ultravox has been acquired by Inworld to accelerate its speech-to-speech research and voice agent capabilities." OpenAI Realtime API: "If you still have a beta Realtime integration, migrate it to the GA interface before moving forward with new work." Synthflow: "Enterprise contracts start at $30,000 annually. Final pricing is scoped around call volume, concurrency, telephony setup, integrations, security needs, and launch support." Only Hume gives a date on the pages we read.
Is Hume EVI shutting down?
Yes. Hume's EVI overview says: "Hume is sunsetting the TTS and EVI APIs. Access ends November 13, 2026 at 12:01 a.m. EST, and both APIs remain fully supported until then. Any data in your account is permanently deleted after November 13, 2026." Plan a move off EVI before that date; this directory lists 15 other voice agent products with their prices and features.
How do voice agent platforms charge?
In different units, so the prices do not rank: per minute (Pipecat + Pipecat Cloud, LiveKit Agents + LiveKit Cloud, Vapi, Cartesia Managed Agents, Ultravox Realtime, OpenAI Realtime API, Retell AI, Twilio ConversationRelay, Bland AI, ElevenLabs Agents, Deepgram Voice Agent API and Amazon Connect Customer); per agent or agent session minute (LiveKit Agents + LiveKit Cloud and Pipecat + Pipecat Cloud); per audio, speech or text token (OpenAI Realtime API, Gemini Live API and Amazon Nova 2 Sonic); plan and contract prices (Vapi, Retell AI, Bland AI, Synthflow, ElevenLabs Agents, Cartesia Managed Agents, LiveKit Agents + LiveKit Cloud and Ultravox Realtime); platform fees and pass-through (Vapi, Bland AI and Pipecat + Pipecat Cloud); numbers, concurrency and add-ons (Vapi and Retell AI). Every price line is listed in the vendor's own unit with its quote, and what a per-minute rate covers differs by vendor.
Which voice agent platforms provide phone numbers?
7 products state it provides or sells numbers on its own pages: Vapi, Retell AI, Bland AI, Synthflow, Cartesia Managed Agents, LiveKit Agents + LiveKit Cloud and Pipecat + Pipecat Cloud. Others connect over SIP or your own carrier; see the telephony columns for each vendor's words.
Which voice agent frameworks are open source?
LiveKit Agents + LiveKit Cloud (Apache-2.0 (framework); turn detection models under LiveKit Model License) and Pipecat + Pipecat Cloud (BSD 2-Clause). Each also has a hosted cloud priced separately.
How many languages does ElevenLabs Agents support?
Its own overview says two things: "Select from 5k+ voices across 31 languages with customization options" and "A low-latency Text to Speech (TTS) model across 5k+ voices and 70+ languages" Both are shown here as stated.
Is this a benchmark or a ranking of voice agent platforms?
No. Every value is a first-party statement copied from the vendor's own pages, with the quote and the link. Nothing here measures latency or call quality; latency figures appear only under a Vendor-stated label, and prices in different units are never converted or compared.
What do Not stated, Vendor-stated and inferred mean?
Not stated means the pages we read did not say; it is not evidence the feature is missing. Vendor-stated marks a latency claim the vendor makes about itself, unverified. Inferred marks our reading: a plan name taken from a pricing table's column order or a card's position, or a feature backed only by a docs navigation label.
Last verified from each vendor's own pricing pages, docs and license files; every quote was string-matched against the page that day. This is not a benchmark. Prices are shown in the vendor's own unit and change without notice, so check the linked source before you commit.
Snapshot of October 19, 2026 and October 20, 2026: 16 products, 59 price lines, 205 quotes, 4 changes, 1 considered and not included.
Newsletter
Know when a voice platform changes
One email per week. New model releases, agent tooling changes and what they mean for what you build.