LiveKit Pricing 2026: Build, Ship and Scale Plans, Per-Minute Model Costs, and a Real Monthly Bill
As of October 25, 2026, LiveKit Cloud is free on Build, from $50 a month on Ship and from $500 on Scale. Ship includes 5,000 agent minutes, one free US number and 100 inbound phone minutes, then $0.01 a minute each, with the AI models billed per minute on top. A 1,200-minute month of AI receptionist calls on Ship comes to about $74 to $110, and a HIPAA BAA starts at Scale.
How much does LiveKit cost? The short answer
As of October 25, 2026, LiveKit Cloud has four plans: Build is free, Ship starts at $50 a month, Scale at $500 a month, and Enterprise is custom. Ship includes 5,000 agent session minutes, one free US phone number and 100 inbound phone minutes, then charges $0.01 a minute for each. The AI models (speech-to-text, the language model and the voice) are billed per minute on top, from well under a cent to $0.03 a minute for the voice alone. A 1,200-minute month of AI receptionist calls on Ship comes to about $74 to $110.
Every number on this page comes from LiveKit's own pricing page and its per-model price lists, read October 25, 2026. LiveKit is a framework and hosting platform for developers who build voice agents, not a finished receptionist product, so the bill depends on the models your builder picks. That is why this page prices three different stacks.
LiveKit Cloud plans side by side
| What you get | Build | Ship | Scale |
|---|---|---|---|
| Monthly price | $0 | From $50 | From $500 |
| Agent session minutes | 1,000 | 5,000, then $0.01 a minute | 50,000, then $0.01 a minute |
| Concurrent agent sessions | 5 | 20 | Up to 600 |
| Agents deployed | 1 | 2 | 4 |
| Inference credits for AI models | $2.50 (about 50 minutes) | $5 (about 100 minutes), then model prices | $50 (about 1,000 minutes), then discounted model prices |
| US local phone number | 1 free | 1 free, then $1 a month each | 1 free, then $1 a month each |
| US local inbound minutes | 50 | 100, then $0.01 a minute | 1,000, then $0.01 a minute |
| Toll-free numbers | Not listed | $2 a month, $0.02 a minute | $2 a month, $0.02 a minute |
| Session recordings | 1,000 minutes | 5,000, then $0.005 a minute | 50,000, then $0.005 a minute |
| HIPAA (signed BAA) and SOC 2 reports | No | No | Yes |
Enterprise adds volume pricing (inference included), SSO, a shared Slack channel and a support SLA, priced by LiveKit's sales team.
Free calculator
What are missed calls costing you?
Run the 60-second Missed-Call Cost Calculator — most service businesses find five figures a year leaking to voicemail.
Calculate my missed-call cost →Free, no email required to run it.
Is LiveKit free?
The software is. LiveKit's FAQ says its Agents framework and media server are both open source, so a developer can run them on their own servers and pay only for the servers, the phone carrier and the AI models directly.
LiveKit Cloud's free Build plan is a testing plan. It includes 1,000 agent minutes but only 50 inbound phone minutes, about 16 three-minute calls, and the pricing table lists no overage rate for Build. A business that wants an agent answering real calls starts at Ship, $50 a month.
What the AI models cost per minute
LiveKit Inference lets one API key call the speech, language and voice models, and LiveKit publishes a per-minute price for each. These are a few of the common picks on Build and Ship, as listed October 25, 2026:
| Layer | Model | Per minute |
|---|---|---|
| Language model | OpenAI GPT-5 mini | $0.0011 |
| Language model | OpenAI GPT-5.6 Terra | $0.0081 |
| Language model | Google Gemini 3.5 Flash | $0.0061 |
| Speech-to-text | Deepgram Nova-3 (monolingual) | $0.0048 |
| Speech-to-text | Deepgram Flux | $0.0065 |
| Voice (text-to-speech) | Inworld Realtime TTS 1.5 Mini | $0.0090 |
| Voice (text-to-speech) | Deepgram Aura-2 | $0.0180 |
| Voice (text-to-speech) | Cartesia Sonic 3.6 | $0.0300 |
| Speech-to-speech | OpenAI GPT Realtime 2.1 | $0.0676 |
The pattern in that table is the useful part: the voice is usually the most expensive line. On a Cartesia voice, text-to-speech is two thirds of the model cost per minute. Picking the voice is the biggest pricing decision your builder makes, bigger than picking the language model. Scale lowers several of these rates (Cartesia Sonic 3.6 drops to $0.0225, Deepgram Flux to $0.0057).
A worked bill: 1,200 minutes of AI receptionist calls on Ship
Take a service business whose AI receptionist answers 400 calls a month on one local number, averaging 3 minutes each, so 1,200 inbound minutes. That is the same month used in our voice AI pricing comparison, so the totals line up.
| Line item | Math | Monthly cost |
|---|---|---|
| Ship plan | Includes 5,000 agent minutes, 5,000 recording minutes, 1 number | $50.00 |
| Inbound phone minutes | (1,200 − 100 included) x $0.01 | $11.00 |
| Stack A: GPT-5 mini + Nova-3 + Inworld 1.5 Mini | 1,200 x $0.0149, minus $5 credits | $12.88, total about $74 |
| Stack B: GPT-5.6 Terra + Deepgram Flux + Cartesia Sonic 3.6 | 1,200 x $0.0446, minus $5 credits | $48.52, total about $110 |
| Stack C: OpenAI GPT Realtime 2.1 (speech-to-speech) | 1,200 x $0.0676, minus $5 credits | $76.12, total about $137 |
At this size the $50 plan fee is the biggest single line on Stack A, and the 1,200 agent minutes and recordings fit well inside what it includes. That changes past 5,000 minutes a month, where each extra minute adds $0.01 for the agent session plus $0.01 for the phone line, before the models.
For comparison, on the same 1,200-minute month: ElevenLabs, Vapi, Retell and Bland land between about $111 and $184 (vendor pages read October 8, 2026, in our voice AI comparison), Telnyx's Voice AI Agents about $82 to $102 (read October 23), and Amazon Connect about $49 (read October 21). LiveKit with a budget voice is at the low end of the hosted options; with a premium voice it is in the middle.
What the LiveKit pricing page leaves out
- The free plan has no phone overage. 50 inbound minutes is the whole allowance on Build, so it cannot run a real receptionist line.
- HIPAA starts at $500 a month. A signed BAA, SOC 2 Type II reports and region pinning are on Scale and Enterprise only. A dental, medical or therapy office that needs a BAA for its phone agent cannot use Ship, so its floor is $500 a month before usage.
- Testing will start to bill. Text and audio simulations, which test the agent against fake callers, are free on Ship and Scale through October 31, 2026, then $0.02 per text turn and $0.05 per audio turn. A 200-turn audio test suite run every week would be about $43 a month. Build includes 350 text and 60 audio turns.
- Turn detection is free only when LiveKit hosts the agent. End-of-turn detection and adaptive interruption are free for agents on LiveKit Cloud. A self-hosted agent gets a development allowance, then pays $0.50 and $1.20 per 1,000 requests on Ship and Scale.
- Keeping your Twilio number costs extra. LiveKit's Twilio connector is $0.004 a minute on Ship, on top of Twilio's own inbound rate. Using your own SIP trunk instead (Telnyx, for example) is included up to 5,000 minutes on Ship, then $0.004 a minute.
- Concurrency caps. Ship allows 20 calls at once on the agent and 20 at once on LiveKit Inference. That is plenty for one location, but a busy outbound campaign or a storm-day call spike across several locations can hit it.
Bring your own phone line or not?
On the 1,200-minute month, LiveKit's own number costs $11 in inbound minutes. Keeping a Twilio number through the connector is about $15 for the minutes (1,200 x $0.004 on LiveKit plus $0.0085 on Twilio, Twilio's rate as read October 7 in our Twilio pricing breakdown) plus $1.15 for the number. A Telnyx SIP trunk is about $7.24 (1,200 x $0.0052 plus the $1 number, from our Telnyx pricing page), with the SIP minutes inside Ship's 5,000. The difference is a few dollars a month either way. The reason to bring your own line is keeping a number your customers already know.
LiveKit vs Vapi: which one costs less?
Vapi charges a $0.05-a-minute hosting fee plus the models at cost, which put its 1,200-minute month at about $127 to $184 on the components it listed on October 8, 2026. LiveKit charges a flat $50 for the first 5,000 minutes instead of a per-minute fee, so at 1,200 minutes the platform part is about $0.04 a minute, and it falls as volume grows. The models cost about the same on either one. The trade is that Vapi gives a builder more of a dashboard to configure an agent in, while LiveKit expects the agent to be written in code with its framework. Our Vapi vs Bland vs Retell comparison covers the dashboard-first platforms.
Is LiveKit worth it?
For a business with a developer or a builder, yes: it is one of the cheapest ways to run a hosted voice agent at small volume, the code is open source, and the agent is not locked to one model or voice vendor. For an owner who wants to set up a receptionist themselves this afternoon, no: LiveKit is a toolkit, and the time to build and test the agent is the real cost.
Who can build a LiveKit voice agent for my business?
You can hire a developer who knows the LiveKit Agents framework, or a studio that builds the whole system and hands it over. Ask which voice they plan to use, because it moves this bill by more than any other choice, and whether you need a BAA, because that sets the plan.
SuperDupr is one of those studios. We build AI voice agents for service businesses, and our published price for a voice agent build is $5,000 to $10,000, with $200 to $2,000 a month ongoing depending on volume. Most AI builds go live in 2 to 4 weeks, and the usage bills (the platform, the phone line, the models) stay in your name at the vendors' own rates. We do not resell LiveKit. The first step is a free 30-minute strategy session, booked from that page.
Sources
- LiveKit pricing, livekit.io/pricing, read October 25, 2026: Build, Ship, Scale and Enterprise plans; agent session, recording, telephony, SIP, Twilio and WhatsApp connector, inference credit, simulation, turn detection and adaptive interruption allowances and rates; per-minute LLM, STT and TTS prices for Build/Ship and Scale; FAQ on self-hosting.
- Vapi, Retell, Bland and ElevenLabs bills: vendor pages read October 8, 2026, as published in our voice AI pricing comparison.
- Telnyx Voice AI Agents and SIP rates: telnyx.com, read October 23, 2026, as published in our Telnyx pricing page. Amazon Connect: aws.amazon.com, read October 21, 2026. Twilio: twilio.com, read October 7, 2026.
- Worked bill: SuperDupr's arithmetic on the published rates above, for an assumed 400 calls of 3 minutes each.
Frequently Asked Questions
-
As of October 25, 2026, LiveKit Cloud's Build plan is free, Ship starts at $50 a month, Scale at $500 a month, and Enterprise is custom. Ship includes 5,000 agent session minutes, 1 free US local number and 100 inbound phone minutes, then $0.01 a minute for each, and the AI models are billed per minute on top. A 1,200-minute month of AI receptionist calls on Ship is about $74 to $110.
-
The LiveKit Agents framework and media server are open source, so you can self-host them and pay only for servers, the phone carrier and the AI models. LiveKit Cloud's free Build plan includes 1,000 agent minutes but only 50 inbound phone minutes, with no overage rate listed, so it is for testing. Real phone traffic starts on Ship at $50 a month.
-
On Ship, past the included allowances, an agent session is $0.01 a minute and an inbound call on a US local number another $0.01. The models are extra: as of October 25, 2026, GPT-5 mini is $0.0011 a minute, Deepgram Nova-3 speech-to-text $0.0048 and voices from $0.0090 (Inworld 1.5 Mini) to $0.0300 (Cartesia Sonic 3.6). LiveKit's calculator puts a default stack at about $0.048 a minute when every line bills at overage rates.
-
Yes, on Scale and Enterprise. LiveKit's pricing page lists a signed BAA, SOC 2 Type II reports and region pinning on Scale (from $500 a month) and up, not on Build or Ship.
-
Usually, at small volume. On the same 1,200-minute month, LiveKit Ship comes to about $74 to $110 depending on the voice (read October 25, 2026), while Vapi's listed components put it at about $127 to $184 (read October 8, 2026). Vapi charges $0.05 a minute for hosting; LiveKit charges $50 for the first 5,000 minutes. The models cost about the same on both.
-
SuperDupr, Justin McKelvey's company, builds AI voice agents for service businesses on platforms like LiveKit. The published price for a voice agent build is $5,000 to $10,000, with $200 to $2,000 a month ongoing depending on volume, and most AI builds go live in 2 to 4 weeks. Usage stays billed in your name at the vendors' own rates; SuperDupr does not resell LiveKit.