Contents
Fast answer
Four cents a minute separates these, and it buys a great deal. LiveKit publishes $0.01 per agent session minute, with Build free at 1,000 minutes and 5 concurrent sessions, Ship at $50 a month for 5,000 minutes and 20 concurrent, and Scale at $500 for 50,000 minutes and up to 600 concurrent, plus separate meters for third-party SIP at $0.004 or $0.003 a minute and downstream data at $0.12 or $0.10 a gigabyte. Vapi publishes $0.05 a minute for orchestration with the model, speech-to-text and speech synthesis billed by your own providers at cost, or at nothing to Vapi with your own keys, plus $0.005 per SMS or chat message, 10 concurrent lines included on Build and extras at $10 per line a month. Both exclude the model, speech and telephony. The difference is that LiveKit gives you real-time infrastructure and Vapi gives you a voice agent orchestration layer built on top of that kind of infrastructure.
Facts last verified 2026-08
Quick facts
- LiveKit agent minute
- $0.01, identical to the included rate on Ship and Scale
- LiveKit tiers
- Build free 1,000 min; Ship $50/mo 5,000; Scale $500/mo 50,000
- LiveKit concurrency
- 5 on Build, 20 on Ship, up to 600 on Scale
- Vapi rate
- $0.05/min orchestration, $0.005 per SMS or chat message
- Vapi concurrency
- 10 lines included on Build, extras $10 per line per month
- Vapi compliance
- HIPAA $2,000/mo, Zero Data Retention $1,000/mo
- Both exclude
- Language model, transcription, speech synthesis and telephony
On this page
What this page covers
A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.
- 01Quick verdict
- 02Pricing
- 03What you build
- 04Concurrency
- 05Decision table
- 06Where both are wrong
Why does this matter now?
Both of these are developer products and both let you bring your own providers, so they attract the same buyer and get compared on rate. The rate is the least useful thing about the comparison. What actually differs is how many weeks of engineering sit between signing up and having a phone agent that handles interruptions properly, and that number is much larger on LiveKit.
Internal path
Where to go next from this page
These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.
What you should be able to do after this
- Understand what each rate actually covers
- Price the engineering time on each
- Compare concurrency and deployment limits
- Decide how much stack you want to own
How do you do it, step by step?
1. Be precise about what each rate covers
LiveKit's $0.01 agent session minute covers its real-time media and session handling. Vapi's $0.05 covers voice agent orchestration: turn taking, tool calls, call control. Neither includes the language model, transcription, speech synthesis or telephony, so both need those added before any comparison means anything.
2. Price the engineering, because it dominates
At 1,500 minutes a month the difference between the two platform rates is $60. A fortnight of engineering time is worth far more than that, and LiveKit expects substantially more of it because there is no dashboard where you configure an agent and go live. Be honest about who is doing that work and what else they would be doing.
3. Compare concurrency structures
LiveKit includes 5 concurrent sessions on Build, 20 on Ship at $50 a month and up to 600 on Scale at $500. Vapi includes 10 concurrent lines on Build with extras at $10 per line a month. Model your peak hour on both, because concurrency and not monthly volume is what determines whether callers get through.
4. Check the extra meters on LiveKit
Third-party SIP is $0.004 a minute on Build and Ship and $0.003 on Scale, downstream data is $0.12 a gigabyte on Ship and $0.10 on Scale, and extra phone numbers are $1.00 a month each beyond the free local number. Vapi's equivalent extras are $0.005 per SMS or chat message and $10 per additional concurrent line.
Pick LiveKit when, pick Vapi when
Pick Vapi when you want provider control without building the voice layer. It charges $0.05 a minute for orchestration and passes the model, speech-to-text and speech synthesis through at provider cost, or at nothing to Vapi if you bring your own API keys, plus $0.005 per SMS or chat message. Turn taking, tool calls and call control are handled, which is precisely the part that stalls LiveKit projects, and 10 concurrent lines are included on Build with extras at $10 per line a month.
Pick Vapi also when compliance needs to be a budget line rather than a project. It publishes HIPAA at $2,000 a month and Zero Data Retention at $1,000 a month on either plan, which means you can price the requirement rather than scope it. That is unusual disclosure in this category and it is genuinely useful for a regulated business that needs a number before committing.
Pick LiveKit when you want to own the real-time layer and have an engineer who wants that job. It publishes $0.01 per agent session minute, with Build free at 1,000 minutes and 5 concurrent sessions, Ship at $50 a month for 5,000 minutes and 20 concurrent, and Scale at $500 a month for 50,000 minutes and up to 600 concurrent. For an unusual requirement that a packaged agent layer would fight, that control is worth having.
Pick LiveKit also when self-hosting is a requirement rather than a preference. Both the Agents framework and the media server are open source and can run on your own infrastructure with nothing paid to LiveKit at all. Vapi is a managed service with no equivalent. If data location or a regulatory obligation rules out a hosted platform, that constraint should filter the shortlist before any rate does.
What the four cent difference actually buys
Both rates exclude the same four things: the language model, speech-to-text, text-to-speech and telephony. So the comparison is not $0.01 against $0.05 as all-in figures, it is $0.01 against $0.05 for two different amounts of assembled functionality. LiveKit's minute covers real-time media transport and session handling. Vapi's covers voice agent orchestration on top of that kind of foundation. The four cents is the orchestration layer you would otherwise write.
Both ladders are also flat, which is worth noticing because it changes how you buy. Ship at $50 for 5,000 minutes and Scale at $500 for 50,000 both work out at exactly $0.01 a minute, which is also LiveKit's published overage rate. So the tiers buy concurrent sessions and deployments rather than a discount, and the correct move is to pick the smallest tier whose concurrency covers your peak and let minutes spill into overage, because overage costs precisely what included minutes cost.
Then add LiveKit's remaining meters, which have no Vapi equivalent and are easy to omit from a model. Third-party SIP is $0.004 a minute on Build and Ship and $0.003 on Scale. Downstream data is $0.12 a gigabyte on Ship and $0.10 on Scale. Extra phone numbers are $1.00 a month each beyond the free local number. Bandwidth in particular has no intuitive relationship to anything on a business dashboard and routinely surprises on a first real month.
The inference credits are a starter allowance and not a budget: $2.50 on Build, $5 on Ship and $50 on Scale. They cover early testing and will not cover a production month, after which inference bills per model at provider rates. On 1,500 minutes a month the platform difference between these two is about $60, which is the least significant number in the comparison once engineering time enters it.
Where both of these are the wrong answer
Neither is right for a business that wants a working phone agent rather than a platform to build one on. Both are developer products, and there is no dashboard on either where a non-technical person configures an agent and goes live. Bland AI's Start plan has no monthly fee at $0.14 a minute with the model, transcription and premium voices bundled and no token charges, which is a higher rate covering strictly more and requiring no assembly.
Neither is right if you want a forecast without a component table. Both leave you assembling a bill from four or five suppliers that each move with usage. Retell AI publishes an itemised $0.07 to $0.31 a minute including the model and US telephony, with 20 concurrent calls free, which gives you an accurate forecast from the published page plus a visual builder that gets a technical operator a long way without an engineer.
Self-hosting LiveKit is not right to save a platform fee. Running real-time media at scale means capacity planning, network tuning and being on call for something customers hear in real time. It is a serious operational undertaking rather than a weekend project, and the $50 a month Ship tier is not the cost that justifies taking it on. Choose self-hosting to satisfy a requirement, never as an optimisation.
And we are not going to separate these two on call quality, latency or interruption handling, because we have run no controlled tests and neither vendor has published ones we would trust. Both are built by teams that understand real-time voice. The honest way to decide on quality is twenty real calls from real phones on each, listened to by someone who knows what a good call sounds like in your business.
How to check this yourself, and what we did not test
All figures on this page were read from each vendor's own pricing page on 20 August 2026 and the source URLs are listed below. LiveKit's tiers and additional meters came from livekit.com/pricing, noting that livekit.io now redirects there, and Vapi's rates and compliance add-ons from vapi.ai/pricing. Both exclude the model, speech and telephony, so any all-in figure you build from this page requires adding your own provider rates.
We did not measure call quality, latency, interruption handling or barge-in behaviour on either platform, and neither vendor has published tests we would trust. Both are built by teams that understand real-time voice. We also did not measure bandwidth consumption, which is a real gap given that LiveKit bills downstream data at $0.12 or $0.10 a gigabyte and that meter is the least intuitive one here.
Build the same simple agent on both free tiers, LiveKit Build with 1,000 agent minutes and Vapi's pay-as-you-go Build plan, and time how long each takes to reach a call you would put in front of a customer. That duration, measured in days of your engineer's time rather than in cents per minute, is what actually separates these two, and it is the number neither pricing page contains.
At a glance
LiveKit Agents vs Vapi decision table, read from each vendor's own pricing page on 20 August 2026
| Decision factor | LiveKit Agents | Vapi |
|---|---|---|
| Platform rate | $0.01 per agent session minute | $0.05 per minute for orchestration |
| What the rate covers | Real-time media transport and session handling | Voice agent orchestration: turn taking, tool calls, call control |
| Free tier | Build: 1,000 agent minutes, 5 concurrent sessions | Build plan is pay-as-you-go with minutes included to start |
| Paid tiers | Ship $50/mo 5,000 min; Scale $500/mo 50,000 min | Build pay-as-you-go; Scale is an annual contract |
| Concurrency | 5, 20 and up to 600 by tier | 10 lines on Build; extras $10 per line per month |
| Model, speech, telephony | Yours to buy and integrate | Yours to buy; at cost through Vapi or $0 with your own keys |
| Extra meters | SIP $0.004 or $0.003/min; data $0.12 or $0.10/GB; numbers $1.00/mo | $0.005 per SMS or chat message |
| Compliance | Build to your own requirements | HIPAA $2,000/mo, Zero Data Retention $1,000/mo |
| Self-hosting | Framework and media server are open source | Managed service |
| Engineering required | Substantial; there is no configure-and-go dashboard | Significant, but far less than building on infrastructure |
| Call quality | Not separated here; we ran no controlled tests | Not separated here; we ran no controlled tests |
What mistakes should you avoid?
- Comparing $0.01 against $0.05 without noting how much more you build on the cheaper one
- Treating LiveKit's $2.50, $5 or $50 inference credits as an inference budget
- Ignoring LiveKit's bandwidth and SIP meters when modelling the bill
- Choosing either without a developer who will still own it in six months
- Sizing on monthly minutes when LiveKit's overage equals its included rate
- Assuming either includes the language model, speech or telephony, because neither does
FAQ
- Which is cheaper, LiveKit or Vapi?
- LiveKit on the platform rate, at $0.01 an agent session minute against Vapi's $0.05. Neither includes the model, speech or telephony. Once engineering time is counted, LiveKit is usually the more expensive of the two, because it provides infrastructure rather than a voice agent orchestration layer.
- What is the actual difference between them?
- Layer. LiveKit provides real-time media transport and session handling. Vapi provides voice agent orchestration on top of that kind of foundation: turn taking, tool calls and call control. If you use LiveKit you build the orchestration; if you use Vapi you configure it.
- Can I self-host either?
- LiveKit yes, since both the Agents framework and the media server are open source and can run on your own infrastructure with nothing paid to LiveKit. Vapi is a managed service. If self-hosting is a compliance or data-location requirement rather than a cost preference, that decides the comparison outright.
- Which has better compliance options?
- Vapi publishes its add-ons openly at $2,000 a month for HIPAA and $1,000 a month for Zero Data Retention on either plan, which lets you price the requirement. LiveKit's approach is that you build to your own requirements on infrastructure you control, which is more flexible and less immediately budgetable.
Sources & further reading
Done for you
Want this built for your business?
Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.
Free scoping call. You own the code.
Was this page helpful?

