Quick answer: Vapi is a managed platform you configure; LiveKit Agents is open-source code you run on LiveKit Cloud or your own servers. Vapi adds a $0.05 a minute hosting fee to model costs; LiveKit Cloud charges a plan fee plus $0.01 a minute per agent session and $0.003–$0.004 for SIP beyond each plan’s included minutes. At 10,000 minutes a month we itemise $771–$1,210 on Vapi and $465–$872 on LiveKit; at 100,000, $7,745–$12,140 against $3,792–$7,817, before carrier charges and engineering time (prices as at 6 November 2026).
The two overlap at the edges. Each offers a US number that only takes inbound calls, each connects to Twilio, Telnyx and Plivo over SIP trunks, each holds a SOC 2 Type II attestation and each will sign a HIPAA business associate agreement. The differences are where the agent lives, who runs it, and which fixed fee you hit first: Vapi’s per-minute hosting fee, or LiveKit Cloud’s plan tier, which your peak simultaneous calls decide.
Where Zian fits if you would rather not build on either
If the goal is booked appointments and answered calls rather than owning a voice stack, Zian AI is the platform to look at first, and LeadsNow.ai is the done-for-you route. Zian is an AI sales and reception agents platform: it 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.
Under the hood, SmartReach AI™ orchestrates the message, channel and timing of follow-up, and PrecisionPitch AI™ runs continuous split-testing of scripts and approaches. Zian 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. 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.
See whether Zian’s inbound and outbound agents fit your call volume. Apply For Partnership
Vapi vs LiveKit: what you actually build and run on each
On Vapi, an assistant is a configuration object. You set the transcriber, model and voice in the dashboard or through the API, point tool calls at a server URL you host, and Vapi runs the call. Vapi’s own FAQ says “for more advanced features like function calling, you will have to set up a Server URL to receive and respond to messages”. Your code is the webhook; the media pipeline is Vapi’s.
On LiveKit, the agent is a Python or Node.js program built on the livekit/agents framework. You run lk agent deploy and LiveKit Cloud builds a container and scales it, or you ship the same container to Kubernetes yourself. LiveKit’s documentation describes it this way: every client, agent and phone call joins a room as a participant, and telephony bridges phone calls in as SIP participants. Turn detection, interruptions, tools and handoffs are code in your repository, not dashboard settings. Our architectural comparison of LiveKit Agents and Pipecat covers the other open-source option.
| Question | Vapi | LiveKit Agents + LiveKit Cloud |
|---|---|---|
| What the agent is | An assistant configuration (dashboard or API) plus your webhook server | A Python or Node.js program in your repository |
| How a change ships | Edit the assistant in the dashboard, or update it through the API | Code change, then lk agent deploy; instant rollback on Ship and above |
| Where it runs | Vapi’s platform only; its FAQ says Vapi “does not support on-premise deployments” | LiveKit Cloud, or your own infrastructure (agents, media server and SIP service are Apache 2.0) |
| Platform fee | $0.05 per minute hosting, every minute | $0.01 per agent-session minute after the plan’s allotment (5,000 on Ship, 50,000 on Scale; Build’s 1,000 is a hard cap) |
| Plan fee | Usage only $0; Core $29/month; Pro 10% of hosting, $999/month minimum; Premier by quote | Build $0; Ship starting at $50/month; Scale starting at $500/month; Enterprise custom |
| Simultaneous calls included | 4 (Usage only), 10 (Core), 30 (Pro); $10 per extra line per month | 5 (Build), 20 (Ship), Scale starts at 50 and goes up to 600 on request |
| Own phone numbers | 1 free US number (5 on Core, 10 on Pro); inbound only | 1 free US local number, then $1/month local or $2/month toll-free; inbound only |
| Outbound calls | Import a number from Twilio or another supported provider, or bring a SIP trunk | Third-party SIP trunk ($0.004/min Ship, $0.003/min Scale, plus the carrier) |
| Packaged outbound campaigns | Yes: Campaigns (CSV upload, schedule, max concurrency) | No packaged campaign product, by LiveKit’s own account; you write the dialling logic |
| Billing increment | We could not find one published, as at 6 November 2026 | Per second, 10-second minimum per session |
| HIPAA BAA | $2,000/month add-on per organisation, on any package | Scale and Enterprise plans |
| SOC 2 Type II | Yes, covering all customers | Yes; audit reports listed on Scale and Enterprise |
| EU or regional processing | EU region with Premier only | Region pinning on Scale and Enterprise |
How Vapi and LiveKit Cloud charge for a minute
Vapi’s pricing FAQ is explicit: “Usage is your Vapi hosting fee ($0.05/min) plus the model costs for each minute your agents run.” The per-minute cost has “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.” Transport through Vapi’s own telephony or a SIP trunk is listed as free on the Vapi side; carrier minutes are “charged by provider, not Vapi”, and the page lists Twilio at $0.008 inbound and $0.014 outbound, Vonage at $0.00814 and Telnyx at $0.0055 a minute. Success packages sit on top of usage and change support, concurrency and retention, not the per-minute rate. Vapi’s FAQ adds that volume pricing is available as you scale; the figures here are list prices.
LiveKit Cloud splits the minute into separate meters. An agent session costs $0.01 a minute after the plan’s included minutes, third-party SIP costs $0.004 a minute on Ship and $0.003 on Scale after 5,000 or 50,000 included minutes, and LiveKit Inference models are billed per token, per character or per second of audio. LiveKit’s billing guide says agent session time is metered “in increments of 1 second with a 10-second minimum per session”, and SIP and WebRTC sessions carry the same 10-second minimum. That minimum matters for outbound: a call that connects to voicemail and drops after four seconds is billed as ten on both the session and SIP meters.
We could not find a billing increment on Vapi’s pricing page, its pricing and billing documentation, its terms of service or its last 90 days of release notes, as at 6 November 2026. Your carrier’s own rounding applies on top of either platform, because both pass carrier minutes through rather than reselling them on the trunk route.
One more thing changes the model line before you compare it. Vapi’s cost documentation says it transcribes “audio from both the caller and the assistant” and that its estimate therefore includes two audio channels. LiveKit’s calculator assumes 3,000 input tokens, 175 output tokens and 600 spoken characters a minute. The two model ranges below come from two different sets of assumptions, so treat the model rows as each vendor’s own estimate, not as a like-for-like quote.
What do 10,000 minutes a month cost on Vapi vs LiveKit?
The worked bill uses the same three model vendors on both sides: Deepgram for speech-to-text, OpenAI for the language model and Cartesia for the voice, each priced from the range that platform publishes. We assume a peak of 10 simultaneous calls, outbound or inbound over your own SIP trunk, with the carrier left out on both sides because it bills separately on both. On Vapi, Core at $29 covers 10 lines, which is cheaper than Usage only plus six extra lines at $10 each ($60). On LiveKit, Ship covers 20 simultaneous agent sessions.
| Line item at 10,000 min/month | Vapi (Core) | LiveKit Cloud (Ship) |
|---|---|---|
| Plan or package fee | $29 | $50 |
| Platform minutes | 10,000 × $0.05 = $500 | (10,000 − 5,000 included) × $0.01 = $50 |
| SIP / transport fee | $0 (BYO SIP listed as free) | (10,000 − 5,000 included) × $0.004 = $20 |
| Speech-to-text (Deepgram) | $0.00953–$0.00994/min = $95.30–$99.40 | $0.0048–$0.0078/min = $48–$78 |
| Language model (OpenAI) | $0.00774–$0.04523/min = $77.40–$452.30 | $0.00022–$0.03787/min = $2.20–$378.70 |
| Voice (Cartesia) | $0.00689–$0.01294/min = $68.90–$129.40 | $0.03/min = $300 |
| Inference credits | — | −$5 |
| Total, before carrier | $770.60–$1,210.10 | $465.20–$871.70 |
| Optional: recordings and observability | Raw data kept 30 days on Core, no separate charge listed | +$25 recordings (5,000 over allotment × $0.005) +$15 events = $505.20–$911.70 |
At this volume the platform layer is $529 on Vapi and $120 on LiveKit ($160 with recordings and observability events). The model layers overlap: the cheapest LiveKit stack runs on a $0.00022 a minute OpenAI model (GPT-5 nano on LiveKit Inference’s published table), while LiveKit’s Cartesia line is higher than Vapi’s. LiveKit prices voice at 600 characters a minute; Vapi’s cost guide says only that it uses “the number of characters spoken in a typical minute”, without a figure we could find.
What do 100,000 minutes a month cost on Vapi vs LiveKit?
At 100,000 minutes we assume a peak of 40 simultaneous calls. That one number sets both plans. On Vapi, Core plus 30 extra lines costs $329 a month, which beats Usage only plus 36 lines ($360) and Pro. Pro is 10% of the $5,000 hosting fee, so the $999 minimum applies, plus 10 lines beyond its included 30, for $1,099. On LiveKit, Ship’s 20 simultaneous sessions are not enough, so the bill moves to Scale, where concurrency starts at 50.
| Line item at 100,000 min/month | Vapi (Core + 30 lines) | LiveKit Cloud (Scale) |
|---|---|---|
| Plan or package fee | $29 + 30 × $10 = $329 | $500 |
| Platform minutes | 100,000 × $0.05 = $5,000 | (100,000 − 50,000 included) × $0.01 = $500 |
| SIP / transport fee | $0 | (100,000 − 50,000 included) × $0.003 = $150 |
| Speech-to-text (Deepgram) | $953–$994 | $0.0042–$0.0068/min = $420–$680 |
| Language model (OpenAI) | $774–$4,523 | $0.00022–$0.03787/min = $22–$3,787 |
| Voice (Cartesia) | $689–$1,294 | $0.0225/min = $2,250 |
| Inference credits | — | −$50 |
| Total, before carrier | $7,745–$12,140 | $3,792–$7,817 |
| Optional: recordings and observability | Raw data kept 30 days on Core, no separate charge listed | +$250 recordings (50,000 × $0.005) +$150 events = $4,192–$8,217 |
The platform layer is now $5,329 on Vapi against $1,150 on LiveKit ($1,550 with observability), a gap of $3,779 to $4,179 a month. That gap is roughly what you pay Vapi not to own agent code; whether it is worth it depends on engineering time these tables leave out. Our platform-versus-framework crossover calculation loads build, evaluation and on-call time into the framework side and finds the break-even far higher than infrastructure prices alone suggest.
LiveKit publishes its own three-way comparison, dated 6 October 2026, with different model choices from ours. Its rows put Vapi at about $70–$116, $698–$1,156 and $3,492–$5,782 at 1,000, 10,000 and 50,000 minutes, against LiveKit Cloud at about $10–$73, $294–$919 and $1,096–$3,844. LiveKit says its Vapi figures are Vapi’s own pricing-calculator totals with AssemblyAI, OpenAI and Cartesia and no success package; the direction matches ours.
Rather not build, run or price either stack? Zian runs AI receptionist, outbound calling and messaging agents as one platform. Apply For Partnership
The Peak-Line Rule: size your simultaneous calls before you price minutes
The Peak-Line Rule: on both Vapi and LiveKit Cloud, your peak number of simultaneous calls, not your monthly minutes, decides which plan you must buy. Size the peak first, then multiply the minutes.
The rule falls out of the two bills above. On LiveKit, the move from 20 to 40 simultaneous calls forces Ship to Scale. At 100,000 minutes the $450 a month difference in plan fee is more than repaid: Ship’s platform layer would be $1,380 ($50 + 95,000 × $0.01 + 95,000 × $0.004) against Scale’s $1,150, before Scale’s discounted inference rates. On Vapi, every line above the package default is $10 a month, so concurrency adds cost in a straight line while the $0.05 hosting fee keeps scaling with minutes. The same 100,000 minutes dialled in a four-hour window needs far more lines than if spread across twelve hours.
| If your peak simultaneous calls are… | Cheapest Vapi package | Cheapest LiveKit Cloud plan |
|---|---|---|
| 4 or fewer | Usage only ($0) | Build ($0) up to 1,000 agent minutes a month, otherwise Ship |
| 5 to 10 | Usage only + $10 per line above 4 at 5 or 6 ($10–$20); Core ($29) from 7 | Build at exactly 5 within 1,000 minutes; otherwise Ship (from $50) |
| 11 to 20 | Core + $10 per line above 10 | Ship (from $50) |
| 21 to 50 | Core + lines (Pro only if you need its SLAs) | Scale (from $500) |
| 51 to 600 | Core + lines, or Premier by quote | Scale, raising the limit from the dashboard |
| Above 600 | Core + lines, or Premier by quote | Enterprise by quote |
To work out your own peak, our guide to sizing voice agent workers for peak calls converts monthly minutes, call length and calling window into simultaneous calls. Vapi’s concurrency guide states the consequence plainly: when all slots are busy, “you cannot start an outbound call or accept an inbound call until a slot becomes available.”
Concurrency limits on Vapi and LiveKit Cloud
Vapi measures concurrency at the subscription level, and all organisations in a subscription share it. Usage only includes 4 lines, Core 10, Pro 30 and Premier a custom number, and extra lines are $10 each per month. A campaign’s maximum concurrency caps that campaign but does not buy or reserve lines.
LiveKit counts three things separately, and its pricing FAQ explains the difference. Concurrent agent sessions are live conversations (5, 20, or 50 rising to 600 on Scale). LiveKit Inference concurrency limits simultaneous model connections per model type on the same tiers. Agent deployments are distinct agents you host (1, 2, 4). On the free Build plan the included allowance is a hard cap, and deployed agents can cold-start with “up to 10 to 20 seconds of delay” after a quiet period. Cold-start prevention starts on Ship.
Phone numbers, SIP trunks, Twilio and Telnyx on Vapi and LiveKit
Neither vendor’s own phone numbers will dial out today. Vapi’s documentation says “Free Vapi numbers are inbound only. Import a number from a supported telephony provider to make outbound calls,” and that free numbers cannot run Outbound Campaigns or be ported out. LiveKit’s documentation says “LiveKit Phone Numbers currently only supports inbound calling. Support for outbound calls is coming soon,” and its product page adds that international numbers will follow. Both were read on 6 November 2026.
For outbound calls, both run through a SIP trunk you hold. Vapi documents Twilio import plus SIP integrations for Telnyx, Plivo, DIDWW, didlogic, Bitcall, Zadarma and Amazon Chime SDK, and accepts any custom “BYO” trunk. LiveKit publishes setup guides for Twilio, Telnyx, Plivo, Sinch and DIDWW, accepts any SIP provider, and adds connectors that bridge Twilio Programmable Voice and WhatsApp Business calls into rooms. LiveKit’s own pricing notes suggest budgeting about $0.008 a minute for the carrier on top of its SIP fee.
For inbound-only agents the arithmetic shifts. A Vapi number carries no transport fee on the Vapi side, while LiveKit charges $0.01 a minute for US local inbound minutes after 100 included on Ship or 1,000 on Scale, and $0.02 for toll-free. LiveKit’s rental limits are 100 local numbers on Ship and 300 on Scale. Both document DTMF, cold and warm transfers and answering-machine detection.
Can I self-host LiveKit or Vapi?
LiveKit, yes. The livekit/agents and agents-js repositories, the livekit server and the livekit/sip bridge are all published under the Apache License 2.0, read on GitHub on 6 November 2026. The latest agents release we saw was livekit-agents 1.8.11, published 5 November 2026. LiveKit’s pricing FAQ says the framework and media server “are both completely open source and available to run locally or host on your own infrastructure.”
Self-hosting has a price of its own. A self-hosted agent still connected to LiveKit Cloud counts against WebRTC participant minutes instead of agent-session minutes. LiveKit’s hosted turn-detection and adaptive-interruption models are free only for agents deployed to LiveKit Cloud. Self-hosted agents get a development allowance of 7,500 and 40,000 requests a month. Beyond that they pay $0.50 and $1.20 per 1,000 requests on Ship and Scale, or fall back to local models on Build.
Vapi, no. Its FAQ answers “Do you support on-premise deployments?” with “No, Vapi does not support on-premise deployments.” Premier lists “Dedicated SIP available”, which is a network option, not self-hosting. We found no on-premise entry in Vapi’s release notes for the 90 days to 2 November 2026.
Latency: what each vendor actually publishes
Vapi’s FAQ says it is designed for “typically around 800 milliseconds for end-to-end voice processing.” LiveKit’s pricing page says its media transport delivers voice and video worldwide “in under 250ms”, which describes the network leg, not a full turn. LiveKit’s own comparison post says that none of the three platforms it compares (Retell, Vapi and LiveKit) publishes a comparable end-to-end benchmark. LiveKit also publishes a live LLM latency benchmark that replays the same scripted conversation against each model every 10 minutes and reports time to first token and first sentence as medians and 95th/99th percentiles; it measures the model leg, not a full call. These figures measure different things and should not be set side by side. Our explainer on what vendor latency numbers actually measure sets out how to test both on your own calls.
HIPAA, SOC 2 and data residency on Vapi vs LiveKit
| Control | Vapi | LiveKit |
|---|---|---|
| SOC 2 Type II | Attested; covers every customer, report via the trust portal | Audited; reports listed on Scale and Enterprise, artefacts under NDA |
| HIPAA BAA | $2,000/month add-on per organisation, any package; organisation-wide mode, compliant providers only | Signed BAA on Scale and Enterprise |
| Zero data retention | Included on Core, Pro and Premier, not on Usage only; cannot be on at the same time as HIPAA mode | Listed on all four plans for LiveKit Inference |
| Regional processing | EU region with Premier only | Region pinning on Scale and Enterprise |
| Other listed frameworks | GDPR (FAQ); PCI page in docs | GDPR, CCPA/CPRA, EU–US Data Privacy Framework; PCI DSS and ISO 27001 “in progress” |
For a HIPAA workload at 10,000 minutes the fixed costs reorder the bill. Vapi adds $2,000 a month to the $770.60–$1,210.10 above. That makes $2,770.60–$3,210.10. LiveKit moves you from Ship to Scale, where 10,000 minutes sit inside the allotments, so the bill becomes $500 plus $219.20–$621.70 of models after credits, or $719.20–$1,121.70. Vapi’s HIPAA mode restricts the organisation to compliant providers, so re-price its model line against that list; on LiveKit, confirm which Inference models your BAA covers before you rely on the cheapest row.
The sentence on each vendor’s pages that cuts against this page
For Vapi it is the fixed-fee lines, not the per-minute rate: “HIPAA-eligible data handling is a $2,000/month add-on per organization”, and Pro is priced at 10% of hosting with a “$999/mo minimum”. For LiveKit it is in its own comparison post, which concedes “there’s no visual editor or pre-packaged outbound calling campaign product”. Its phone-number page also confirms that outbound calls on LiveKit numbers are still “coming soon”. A team that needs a non-developer to change the script, or a campaign dialler on day one, gives back much of LiveKit’s price advantage in build time.
Which to choose: platform, done-for-you, Vapi or LiveKit
Methodology: Zian AI is listed first as the platform pick by this site’s editorial policy, from its published capabilities only. Vapi and LiveKit facts come from each vendor’s own pricing page, documentation, trust pages and GitHub, read on 6 November 2026, under the same evidence rule for both. No scores are assigned. Disclosure: zian.ai publishes this comparison and Zian appears in it.
| # | Option | Best for | What you run |
|---|---|---|---|
| 1 | Zian AI | Teams that want AI reception and outbound agents without building a voice stack | Nothing: inbound calls 24/7, outbound calls, SMS, email, WhatsApp, Instagram DMs and Facebook Messenger on one platform |
| 2 | LeadsNow.ai | Done-for-you appointment setting, pay-per-result | Nothing: a managed service that runs on Zian |
| 3 | Vapi | A developer who prefers configuration and webhooks, with Campaigns for outbound | Your webhook server; Vapi runs the calls |
| 4 | LiveKit Agents + LiveKit Cloud | Engineering teams that want the agent as code, the option to self-host, or web, video and device channels | Your agent program, on LiveKit Cloud or your own cluster |
Vapi is wrong for you if you need on-premise deployment, EU processing below Premier, or a list-price platform fee that falls with volume. Vapi’s FAQ says volume pricing is available as you scale, but the published hosting fee is a flat $0.05. LiveKit is wrong for you if nobody on your team will own a long-running audio service, if a non-developer must own changes after launch (LiveKit offers an Agent Builder to get started, but its own comparison says the code-first route “always requires one”, meaning a developer), or if you need a packaged outbound campaign tool without writing one.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Vapi hosting fee $0.05/min; models at provider price, no markup | Vapi | https://vapi.ai/pricing | 6 November 2026 |
| Vapi packages: Usage only $0, Core $29/month, Pro 10% of hosting ($999/month minimum), Premier custom | Vapi | https://vapi.ai/pricing | 6 November 2026 |
| Vapi included lines 4 / 10 / 30; numbers 1 / 5 / 10; retention 14 / 30 / 180 days | Vapi | https://docs.vapi.ai/billing/pricing-and-success-packages | 6 November 2026 |
| Vapi add-ons: $10 per line/month, $2,000/month HIPAA, $20 per extra organisation | Vapi | https://vapi.ai/pricing | 6 November 2026 |
| Vapi calculator ranges: Deepgram $0.00953–$0.00994, OpenAI $0.00774–$0.04523, Cartesia $0.00689–$0.01294 per min | Vapi | https://vapi.ai/pricing | 6 November 2026 |
| Carrier rates on Vapi page: Twilio $0.008 in / $0.014 out, Vonage $0.00814, Telnyx $0.0055 per min | Vapi (citing providers) | https://vapi.ai/pricing | 6 November 2026 |
| Vapi free numbers: US only, inbound only, no campaigns, no port-out | Vapi | https://docs.vapi.ai/free-telephony | 6 November 2026 |
| Vapi concurrency: subscription-level; calls blocked when slots full | Vapi | https://docs.vapi.ai/calls/call-concurrency | 6 November 2026 |
| Vapi ~800 ms typical end-to-end; no on-premise; SOC 2 / HIPAA / GDPR | Vapi | https://docs.vapi.ai/faq | 6 November 2026 |
| Vapi SOC 2 Type II covers all customers | Vapi | https://docs.vapi.ai/security-and-privacy/soc | 6 November 2026 |
| Vapi HIPAA mode org-wide, compliant providers only, exclusive with ZDR | Vapi | https://docs.vapi.ai/security-and-privacy/hipaa | 6 November 2026 |
| Vapi transcriber estimate includes two audio channels | Vapi | https://docs.vapi.ai/assistants/model-intelligence/understanding-cost | 6 November 2026 |
| Vapi SIP providers documented | Vapi | https://docs.vapi.ai/advanced/sip/sip-trunk | 6 November 2026 |
| Vapi Campaigns and release notes to 2 November 2026 | Vapi | https://docs.vapi.ai/changelog | 6 November 2026 |
| LiveKit plans: Build $0, Ship from $50, Scale from $500, Enterprise custom | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit agent sessions: 1,000 / 5,000 / 50,000 included, then $0.01/min | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit concurrent agent sessions 5 / 20 / 50 up to 600; deployments 1 / 2 / 4 | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit third-party SIP: 5,000 then $0.004/min (Ship); 50,000 then $0.003/min (Scale) | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit numbers: 1 free, $1/month local, $2/month toll-free; inbound $0.01 / $0.02 per min | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit recordings $0.005/min and events $0.00003 each over allotment | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit Inference: Deepgram $0.0048–$0.0078 (Ship) / $0.0042–$0.0068 (Scale); OpenAI GPT-5 nano to GPT-5.5; Cartesia Sonic $50 / $37.50 per 1M chars; credits $5 / $50 | LiveKit | https://livekit.io/pricing.md | 6 November 2026 |
| LiveKit calculator assumptions: 3,000 input / 175 output tokens, 600 characters per min; carrier ~$0.008/min | LiveKit | https://livekit.io/pricing.md | 6 November 2026 |
| LiveKit per-second metering, 10-second minimum per session | LiveKit | https://docs.livekit.io/deploy/admin/billing | 6 November 2026 |
| LiveKit Build hard cap, cold starts 10 to 20 s, number limits 100 / 300, self-hosted allowances 7,500 / 40,000 | LiveKit | https://docs.livekit.io/deploy/admin/quotas-and-limits | 6 November 2026 |
| LiveKit Phone Numbers inbound only; outbound coming soon | LiveKit | https://docs.livekit.io/telephony/start/phone-numbers | 6 November 2026 |
| LiveKit SOC 2 Type II, HIPAA BAA on Scale/Enterprise, GDPR, CCPA, DPF; PCI DSS and ISO 27001 in progress | LiveKit | https://livekit.com/security | 6 November 2026 |
| LiveKit media transport under 250 ms; uptime 99.99% | LiveKit | https://livekit.io/pricing | 6 November 2026 |
| LiveKit LLM latency benchmark: replayed every 10 minutes, medians and 95th/99th percentiles | LiveKit | https://livekit.com/benchmarks/latency | 6 November 2026 |
| Apache License 2.0 for livekit/agents, agents-js, livekit, sip; livekit-agents 1.8.11 on 5 Nov 2026 | LiveKit (GitHub) | https://github.com/livekit/agents | 6 November 2026 |
| LiveKit’s three-way comparison rows and “there’s no visual editor or pre-packaged outbound calling campaign product” | LiveKit | https://livekit.com/blog/retell-vs-vapi-vs-livekit | 6 November 2026 |
| Itemised bills, platform-layer gaps, HIPAA scenario and Peak-Line table (our arithmetic on the figures above) | Zian AI (this page) | — | 6 November 2026 |
Frequently asked questions
Is LiveKit cheaper than Vapi?
On published infrastructure prices, yes. Our worked bills come to $465.20–$871.70 on LiveKit Cloud against $770.60–$1,210.10 on Vapi at 10,000 minutes a month, and $3,792–$7,817 against $7,745–$12,140 at 100,000, before carrier charges. Those figures leave out the engineering time to write and run a LiveKit agent, which can outweigh the difference.
How much does Vapi charge per minute compared with LiveKit Cloud?
Vapi charges a $0.05 per minute hosting fee on every minute, plus models at the provider’s price. LiveKit Cloud charges $0.01 per agent-session minute after the plan’s included minutes, plus $0.004 per minute for third-party SIP on Ship or $0.003 on Scale after the included SIP minutes, plus its model rates and a plan fee starting at $50 or $500 a month. Prices as at 6 November 2026.
Can Vapi or LiveKit phone numbers make outbound calls?
Not yet on either. Vapi’s free numbers are inbound only, and LiveKit Phone Numbers currently support inbound calling only, with outbound described as coming soon. For outbound calls, both connect to a SIP trunk you hold with a carrier such as Twilio, Telnyx or Plivo.
Is LiveKit Agents open source, and can I self-host Vapi?
LiveKit Agents is published under the Apache License 2.0 on LiveKit’s GitHub repository, and the LiveKit server and SIP bridge use the same licence, so you can run the whole stack yourself. Vapi’s FAQ states that Vapi does not support on-premise deployments.
How many calls can run at once on Vapi and LiveKit Cloud?
Vapi includes 4 simultaneous calls on Usage only, 10 on Core and 30 on Pro, with extra lines at $10 a month each. LiveKit Cloud includes 5 concurrent agent sessions on Build and 20 on Ship, and Scale starts at 50 with increases up to 600 requested from the dashboard.
Which is cheaper for HIPAA calls, Vapi or LiveKit?
On our worked bill at 10,000 minutes a month, LiveKit is. Vapi’s HIPAA add-on is $2,000 a month per organisation on top of usage, while LiveKit includes a signed BAA from the Scale plan, which starts at $500 a month. Vapi’s HIPAA mode allows compliant providers only, so check the model line on both before you sign.
Does LiveKit bill per second?
Yes. LiveKit’s billing guide says agent session time is metered in 1-second increments with a 10-second minimum per session, and SIP and WebRTC sessions carry the same 10-second minimum. We could not find a published billing increment for Vapi as at 6 November 2026.