LiveKit Pricing (2026): Cloud Plans and Cost per Minute
LiveKit's Agents framework is open source and free to self-host. LiveKit Cloud, the hosted version, has a free Build plan with 1,000 agent session minutes, Ship from $50 a month with 5,000 minutes, and Scale from $500 a month with 50,000 minutes; past the included minutes an agent session is $0.01 a minute. The models (speech-to-text, LLM and voice, or one realtime model) and the phone line are billed per minute on top, through LiveKit Inference or by your own providers. Every price below is from livekit.com/pricing, read October 25, 2026. The per-minute and monthly totals are Cozzo's arithmetic from those rates, not a bill we measured.
LiveKit's plan price is the small part. Most of a LiveKit bill is the models and the phone line, billed per minute on top. Price your exact speech-to-text, LLM and voice picks before you compare it with an all-in rate like Ultravox's $0.05 a minute.
LiveKit Cloud is free for 1,000 agent minutes, then $50/mo Ship or $500/mo Scale. What a minute, a call and a month really cost with models and phone lines.
What you pay for on a LiveKit Cloud call
Agent session
$0.01/min past allowance
Time your agent spends in a session on LiveKit Cloud. Build includes 1,000 minutes, Ship 5,000 and Scale 50,000 a month; after that it's $0.01 a minute on Ship and Scale. Concurrent sessions are capped at 5 on Build, 20 on Ship and up to 600 on Scale.
LLM (LiveKit Inference)
$0.0006–$0.0379/min
LiveKit converts token prices to a per-minute figure assuming 3,000 input and 175 output tokens a minute. Its own example, Gemma 4 31B, is $0.0014 a minute; GPT-4o mini is $0.0006, GPT-4.1 mini $0.0015, GPT-5.2 $0.0077 and GPT-5.5 $0.0379. Bring your own provider key instead and the LLM is billed by that provider, not LiveKit.
Speech-to-text
$0.0025–$0.017/min
Billed per minute through LiveKit Inference on Build and Ship: AssemblyAI Universal-Streaming $0.0025, Deepgram Nova-3 $0.0048, Deepgram Flux $0.0065, AssemblyAI Universal-3.6 Pro Streaming $0.0075, OpenAI GPT-Live Transcribe $0.017. Scale gets lower rates on some models (Nova-3 $0.0042).
Voice (text-to-speech)
$0.009–$0.03/min
Billed per character; LiveKit assumes 600 characters a minute. Fish Audio S2 Pro and Inworld TTS 1.5 Mini at $15 per 1M characters work out to $0.009 a minute, Deepgram Aura-2 $0.018 and Cartesia Sonic 3 ($50 per 1M) $0.03. Cartesia Sonic 3 is $0.0225 a minute on Scale.
Or one realtime model
$0.0182–$0.08/min
A speech-to-speech model replaces the three lines above. LiveKit lists Gemini 3.8 Live at $0.0182 a minute, GPT Realtime 2.1 mini $0.0216, GPT-Live 1 $0.05, GPT Realtime 2.1 $0.0676 and Grok Voice 3.0 $0.08. The OpenAI and Gemini figures assume 3,200 audio tokens in and 720 out a minute, with 80% of input cached.
Phone line
$0.004–$0.01/min
LiveKit's own US local number: 1 free number, then $1 a month each; inbound minutes $0.01 after 50 included on Build, 100 on Ship, 1,000 on Scale. US toll-free is $2 a month plus $0.02 a minute. Bring your own SIP trunk (Twilio, Telnyx) and LiveKit charges $0.004 a minute after 5,000 included on Ship ($0.003 after 50,000 on Scale), plus your carrier: Twilio inbound is $0.0085 a minute.
Observability (optional)
About $0.01/min if on
LiveKit's estimate for session recordings plus transcripts, traces and logs (about 100 entries a minute). Recordings are $0.005 a minute and events $0.00003 each after the included allowance (Ship: 5,000 recording minutes and 500,000 entries). Leave it off and it costs nothing.
Turn-taking models
Free on LiveKit Cloud
LiveKit's end-of-turn and adaptive interruption models are free for agents hosted on LiveKit Cloud. Self-hosted agents get a development allowance, then pay $0.50 (turn detection) and $1.20 (interruption) per 1,000 requests on Ship and Scale.
Total per minute
about 2.8¢–4.8¢ a minute
On LiveKit's sample stack. Its own calculator default comes to $0.0479 a minute: agent session $0.01, LLM $0.0014, speech-to-text $0.0075, voice $0.009, observability $0.01 and a US local phone minute $0.01. Inside your plan's included session minutes and with observability off, the same stack is about $0.0279. Premium voices or bigger LLMs push it up: Cartesia Sonic 3 alone adds 2.1¢ a minute over the $0.009 voice.
What a month costs on LiveKit's own rates
500 min/month
- Ship + LiveKit number
- $63
- Ship + your Twilio SIP
- $64
- Ship + GPT Realtime 2.1 mini
- $65
- Scale + LiveKit number
- $509
1,000 min/month
- Ship + LiveKit number
- $77
- Ship + your Twilio SIP
- $78
- Ship + GPT Realtime 2.1 mini
- $81
- Scale + LiveKit number
- $518
2,500 min/month
- Ship + LiveKit number
- $119
- Ship + your Twilio SIP
- $117
- Ship + GPT Realtime 2.1 mini
- $128
- Scale + LiveKit number
- $560
5,000 min/month
- Ship + LiveKit number
- $189
- Ship + your Twilio SIP
- $183
- Ship + GPT Realtime 2.1 mini
- $207
- Scale + LiveKit number
- $630
10,000 min/month
- Ship + LiveKit number
- $378
- Ship + your Twilio SIP
- $385
- Ship + GPT Realtime 2.1 mini
- $415
- Scale + LiveKit number
- $769
25,000 min/month
- Ship + LiveKit number
- $947
- Ship + your Twilio SIP
- $991
- Ship + GPT Realtime 2.1 mini
- $1,039
- Scale + LiveKit number
- $1,188
50,000 min/month
- Ship + LiveKit number
- $1,894
- Ship + your Twilio SIP
- $2,001
- Ship + GPT Realtime 2.1 mini
- $2,079
- Scale + LiveKit number
- $1,885
Cozzo's arithmetic from LiveKit's published rates. Each column is the plan fee, agent session minutes past the plan's allowance at $0.01, phone minutes past the allowance, and LiveKit's sample models (LLM $0.0014 + speech-to-text $0.0075 + voice $0.009 = $0.0179 a minute) or GPT Realtime 2.1 mini at $0.0216. The Twilio column adds a $1.15 number and $0.0085 a minute inbound, plus LiveKit's $0.004 SIP fee after 5,000 minutes. Not included: observability, the $5 (Ship) or $50 (Scale) of inference credits, and Scale's lower model rates. Ship and Scale are "starting at" prices. Want your own volume and carrier? Try the cost calculator.
Packages and add-ons
Build (free)
$0/mo
1,000 agent session minutes, 5 concurrent sessions, 1 agent deployment, 1 free US number with 50 inbound minutes, 1,000 third-party SIP minutes and $2.50 of inference credits (about 50 minutes, LiveKit says). No credit card. LiveKit's table lists no overage rates for Build, so treat it as a testing plan.
Ship
From $50/mo
5,000 agent session minutes then $0.01 a minute, 20 concurrent sessions, 2 agent deployments, cold start prevention, instant rollback, 20 custom voices, $5 of inference credits, 100 inbound minutes on a free US number, team collaboration and email support.
Scale
From $500/mo
50,000 agent session minutes, up to 600 concurrent sessions (starts at 50), 4 agent deployments, $50 of inference credits and discounted model rates, region pinning, role-based access, SOC 2 Type II reports and a HIPAA BAA. It's the lowest LiveKit plan with HIPAA.
When Scale beats Ship
About 49,100 min/month
Cozzo's arithmetic on the sample stack with LiveKit numbers: past 5,000 minutes Ship costs about $0.0379 a minute all-in against $0.0279 on Scale (until its 50,000 included minutes run out), so Scale's extra $450 is paid back at roughly 49,100 minutes a month, before Scale's lower model rates. Need HIPAA? Scale is required anyway.
Cozzo's worked example: a 3-minute call
$0.08–$0.14 per call
On Ship with the sample stack and a LiveKit number: about $0.084 while you're inside the 5,000 included session minutes and $0.114 after, or $0.144 with observability turned on. With your own Twilio SIP trunk past the allowances it's about $0.121. Swap in GPT Realtime 2.1 mini and it's about $0.125 past the allowance.
Self-hosting
Free software
The LiveKit Agents framework and media server are open source, so you can run both on your own servers and pay only for hosting, model providers and your carrier. Self-hosted agents connected to LiveKit Cloud count as WebRTC participant minutes (150,000 included on Ship, then $0.0005 a minute) and pay for turn detection after the development allowance.
How it compares on published pricing
LiveKit Agents- Starting price
- Free; Ship from $50/mo
- Best for
- Developers building on open-source realtime infrastructure
Pipecat- Starting price
- Free (self-host); Pipecat Cloud from $0.01/min
- Best for
- Python developers who want full control of the voice pipeline and any model provider
Vapi- Starting price
- $0 to start; $0.05/min Vapi hosting, models billed pass-through at cost
- Best for
- Developers who want to choose their own LLM, transcriber, voice and telephony provider
Ultravox- Starting price
- $0.05/min (first 30 min free)
- Best for
- Developers wanting a speech-native model at a flat per-minute rate
Retell AI- Starting price
- $0 platform fee; $0.07–$0.31/min for voice agents
- Best for
- Teams that want a no-code flow builder plus an API with itemized per-minute pricing
OpenAI Realtime API- Starting price
- gpt-realtime-2.1-mini: audio $10 in / $20 out per 1M tokens
- Best for
- Developers building speech-to-speech voice agents directly on OpenAI models
ElevenLabs Agents- Starting price
- Agents: $0.08 per minute
- Best for
- Builders who want ElevenLabs' voices and speech models in one agent pipeline
Pricing explained, line by line:Vapi pricing →Ultravox pricing →Retell AI pricing →OpenAI Realtime API pricing →ElevenLabs Agents pricing →
Cozzo-tested · Round 4 · October 22, 2026
AI receptionist benchmark: 8 voice agents on 5 real-call scenarios
LiveKit wasn't tested as a LiveKit Cloud product: Cozzo ran its own self-hosted LiveKit agent (Deepgram nova-3, gpt-4o-mini, ElevenLabs Flash v2.5). It passed 11 of 15 Round 4 calls with a median reply of 2.00 seconds, and passed none of the three unknown-insurance calls because it said it couldn't take messages; that was our own prompt and setup, not a LiveKit Cloud agent. Here's how every platform did on the same dental-office calls. Each chip shows calls passed out of 3. Fastest median reply first.
ElevenLabs · 13/15 · 1.06 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
Ultravox · 14/15 · 1.27 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
Millis AI · 12/15 · 1.31 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
Retell AI · 12/15 · 1.38 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
OpenAI Realtime · 12/15 · 1.71 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
Vapi · 9/15 · 1.90 s‡ median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
Bland AI · 10/15 · 1.94 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
LiveKit · 11/15 · 2.00 s median reply
- Book a new patient
- Reschedule
- Unknown insurance
- Interruption
- Dental emergency
‡ Vapi’s 1.90 s median is end-to-end on an untuned default setup, not comparable to Vapi’s published figures, which leave out endpointing and network time. † Callback taken, but the number was read back wrong at least once. * Reply heard each time, but its content wasn’t checked against the vendor’s own transcript. 3 calls per scenario, one prompt, English only: small samples, so treat close numbers as ties.
Cozzo tested: reply time on real phone calls
Median time from the end of the caller’s speech to the agent’s first reply, on each vendor’s default model and voice (Oct 20, 2026). Places are out of 8 platforms tested.
- Retell AI: 1.04 s (tied 1 of 8, 11 calls)
- ElevenLabs Agents: 1.04 s (tied 1 of 8, 11 calls)
- Ultravox: 1.18 s (3 of 8, 11 calls)
- LiveKit Agents: 1.52 s (tied 5 of 8, 11 calls)
- OpenAI Realtime API: 1.52 s (tied 5 of 8, 11 calls)
- Vapi: 1.56 s (7 of 8, 11 calls)
Not tested yet: Pipecat.
See the full phone-call benchmark → Small samples, so treat close numbers as ties.
What to check before you choose
- Price the models first. LiveKit Cloud's own fees (agent session $0.01 a minute, a phone minute $0.01) are often less than half the bill. Your speech-to-text, LLM and voice picks decide most of the rest: the voice alone ranges from $0.009 to $0.03 a minute on LiveKit Inference.
- Use the included minutes. Ship's 5,000 agent session minutes cover more than 1,600 three-minute calls a month. Inside them a call on LiveKit's sample stack is about 2.8¢ a minute; after them it's about 3.8¢.
- Pick the phone line on volume. LiveKit's own US number is free (one per project) and $0.01 a minute after the included inbound minutes. Your own SIP trunk is $0.004 a minute on LiveKit after 5,000 minutes, plus your carrier (Twilio inbound $0.0085), so it rarely saves money unless you already have numbers there.
- Keep observability for testing. Recordings and traces add about $0.01 a minute once you pass the included allowance. Turn them on while you tune the agent, then sample a share of calls in production.
- Watch the concurrency cap. Build allows 5 calls at once and Ship 20. If your phones get a lunch-hour rush, check you won't hit the cap before you pick a plan; Scale starts at 50 and goes up to 600.
- Bring your own keys if you have discounts. LiveKit Inference bills models through one account, but external models are billed by the provider directly with no LiveKit charge for the LLM. That helps if you already have negotiated OpenAI, Deepgram or ElevenLabs rates.
- Measure on real calls. Cozzo's self-hosted LiveKit agent replied in a median 1.52 seconds over 11 phone calls on Oct 20, end to end. Your model picks change both speed and cost, so test a few calls before you commit to volume.
Questions people ask
How much does LiveKit cost per minute?
LiveKit Cloud charges $0.01 a minute per agent session after your plan's included minutes, plus the models and phone line. LiveKit's calculator default comes to $0.0479 a minute: session $0.01, LLM $0.0014, speech-to-text $0.0075, voice $0.009, observability $0.01 and a US local phone minute $0.01. Inside the included session minutes and with observability off it's about $0.028.
Is LiveKit free?
The LiveKit Agents framework and media server are open source and free to self-host. LiveKit Cloud's Build plan is $0 a month with 1,000 agent session minutes, 5 concurrent sessions, one free US number with 50 inbound minutes and $2.50 of inference credits (about 50 minutes). LiveKit lists no overage rates for Build, so production traffic usually means Ship.
What does the LiveKit Ship plan include?
Ship starts at $50 a month: 5,000 agent session minutes then $0.01 a minute, 20 concurrent sessions, 2 agent deployments, cold start prevention, instant rollback, 20 custom voices, $5 of inference credits, one free US number with 100 inbound minutes, and email support.
When is LiveKit Scale worth $500 a month?
On LiveKit's sample model stack with LiveKit phone numbers, Cozzo's arithmetic puts the break-even with Ship at about 49,100 minutes a month, before Scale's lower model rates. Scale is also the lowest plan with a HIPAA BAA, region pinning, role-based access and SOC 2 Type II reports, so medical practices need it regardless of volume.
How much do phone calls cost on LiveKit?
A US local number from LiveKit is free (one per project), then $1 a month per number; inbound calls are $0.01 a minute after 50 (Build), 100 (Ship) or 1,000 (Scale) included minutes. Toll-free numbers are $2 a month and $0.02 a minute. With your own SIP trunk, LiveKit charges $0.004 a minute after 5,000 included minutes on Ship, and your carrier bills its own rate on top.
Can I use my own OpenAI, Deepgram or ElevenLabs keys with LiveKit?
Yes. LiveKit's pricing rules say models billed outside LiveKit Inference cost $0 through LiveKit and are billed by the model provider directly. You still pay LiveKit Cloud for agent session minutes, telephony and any observability.
Does LiveKit charge for turn detection?
Not for agents hosted on LiveKit Cloud: its end-of-turn and adaptive interruption models are free there. Self-hosted agents get a development allowance, then pay $0.50 per 1,000 turn-detection requests and $1.20 per 1,000 interruption requests on Ship and Scale.
Is LiveKit cheaper than Pipecat Cloud?
They're close on hosting. Pipecat Cloud charges $0.01 per active minute on its agent-1x profile with unlimited concurrency, and $0.018 a minute for PSTN dial-in or dial-out; models are billed by your providers. LiveKit Cloud includes 5,000 agent minutes on the $50 Ship plan, then $0.01 a minute, with phone minutes at $0.01. See our LiveKit vs Pipecat comparison for the full picture.
Is LiveKit cheaper than Vapi or Ultravox?
Often, if you pick low-cost models. LiveKit's sample stack is about 2.8–4.8¢ a minute including a phone minute. Vapi charges $0.05 a minute hosting plus your providers, and Ultravox a flat $0.05 a minute including its model, plus SIP. LiveKit asks more of you, though: you write and deploy the agent in code.
Has Cozzo measured what a LiveKit call costs?
Not as a LiveKit Cloud bill. Our LiveKit tests used a self-hosted LiveKit agent (Deepgram nova-3, gpt-4o-mini, ElevenLabs Flash v2.5), so the dollar figures here are LiveKit's published rates and our arithmetic. We did measure speed: a median reply of 1.52 seconds over 11 phone calls on Oct 20, and 2.00 seconds in our Round 4 receptionist calls, where it passed 11 of 15.
Ready to look closer? See LiveKit Agents’ full listing and price history → Or weigh the options in LiveKit Agents Alternatives Compared.
Not sure which fits? Answer a few quick questions in the finder →
Prices from vendors' official pricing pages, checked October 25, 2026. Not sponsored; order isn't a ranking.