Voice AI
GoHighLevel Voice AI cost: what a call actually bills
GoHighLevel Voice AI now runs on two architectures priced very differently. Speech-to-speech is a single flat rate, $0.10 a minute on Gemini 3.1 Flash Live Preview or $0.20 on GPT Realtime 2.1, free until 6 September 2026. The older cascading pipeline unbundles into four meters and lands between $0.060 and $0.215 a minute. Telephony bills separately on both.
Research-led reviewThe author has not used this product. Built from public sources.What this means
The author has not used this product. Everything here comes from six public sources: the maker's own site, its pricing and refund terms, its changelog, its support forum, counted complaints across recent public reviews, and a like-for-like comparison against the nearest alternatives. We never claim experience we do not have.
Which badge a page carries is enforced in the site build, not just in editing. A page cannot use first-hand language unless it has declared itself a real-use review, so the badge and the writing cannot drift apart. The full method
We earn a commission if you buy through links on this page, at no extra cost to you.How this works
You pay exactly what you would pay going direct. The maker pays us a share of their margin, not a surcharge on you.
We are never paid for a positive verdict, no maker sees a review before it publishes, and we publish verdicts recommending against products we could earn from. Commission rates never influence which products we cover or how we rank them. Full disclosure
The short version
- Two architectures now. Speech-to-speech is one flat rate, the older pipeline is four meters.
- The s2s free period ended on 6 September 2026, and billing started automatically.
- On the cascading pipeline, the voice you choose changes the floor by nearly four times.
- AI Employee Unlimited at $97 covers the AI but never the phone minutes, on either.
Affiliate link. We earn a commission if you subscribe, at no extra cost to you. 30 day free trial through HighLevel's Bootcamp offer, card required.
Two architectures, two prices
Almost everything written about GoHighLevel Voice AI describes only the older setup, because until recently that was the only one. A cascading pipeline chains three services together: one converts speech to text, a language model decides what to say, and a third turns that back into speech. Each link bills separately, which is why the price is a stack rather than a rate.
Speech-to-speech collapses all of that into a single model that hears and speaks directly. That removes the hand-off between links, which is the delay that makes an AI agent sound like it is waiting for its turn to talk. It is also priced as one flat per-minute figure, which makes forecasting far simpler.
| Model | Rate | Notes |
|---|---|---|
| Gemini 3.1 Flash Live Preview | $0.10 / min | The cheaper of the two, and the default choice for volume. |
| GPT Realtime 2.1 | $0.20 / min | Twice the price. Worth a side-by-side before you commit a client to it. |
The free period is the part worth understanding properly. Agents running on s2s were free to 6 September 2026, and after that date they simply started billing at the model rate. There was no prompt, no opt-in and no pause. If you set up a voice agent during that window and forgot about it, it has been charging since.
That pattern, a free feature that quietly becomes a metered one, is the same shape as several other charges on this platform. We keep a running list of them on GoHighLevel upgrades.
| Question | Answer |
|---|---|
| What speech-to-speech changes | One model hears and speaks, instead of three services passing text between them. That removes the hand-off delay that makes a cascading agent sound like it is waiting for its turn. |
| What you get for it | Faster responses, more natural conversation, emotional awareness, and language switching mid-sentence rather than only between calls. |
| How it is priced | One flat per-minute rate for the model. There is no separate engine fee, speech fee and token bill to add up, which makes forecasting far easier than the cascading pipeline. |
| What it still does not cover | Telephony. The phone call itself bills at Twilio rates on both architectures, and no AI plan has ever included it. |
| The catch on the free period | It ended on 6 September 2026. Agents left running on s2s after that date started billing at the model rate automatically, with no prompt and no opt-in. |
On price alone, s2s at $0.10 is roughly double the $0.060 floor of the cheapest cascading setup, and roughly half the $0.215 of the most natural-sounding one. So it is not a saving if you were running the cheap voice, and it is a saving if you were running the good one. The difference is that $0.10 is the whole bill rather than the first line of it.
The cascading pipeline, priced as a stack
The older architecture is still there, still the default on existing agents, and still what nearly every article about Voice AI pricing describes. Four meters run at once, which is why quoting it as a single per-minute figure never works.
| Component | Rate | Notes |
|---|---|---|
| Voice engine | $0.045 / min | The base charge, on every call, whichever voice you pick. |
| Speech, OpenAI or Cartesia | $0.015 / min | Cheapest option. Floor of about $0.060 a minute all in. |
| Speech, ElevenLabs V2.5 | $0.035 / min | Better voices. Floor of about $0.080 a minute. |
| Speech, ElevenLabs V3 | $0.170 / min | Best voices, and nearly four times the floor at about $0.215 a minute. |
| Language model tokens | Per model | From cheap small models to a few dollars per million tokens on frontier ones. |
| Telephony | Twilio rates | Billed separately. The AI plan never covers the phone call itself. |
What a real call costs
Add telephony at Twilio rates and a typical inbound booking call is still comfortably under a dollar and change. That is the honest headline: for the job of answering, qualifying and booking, the per-call economics are not the problem.
Where it gets expensive is volume across clients. Because AI Employee is billed per sub-account, running voice across ten clients means ten subscriptions plus every minute. That maths is on the AI pricing page, and it is the part worth modelling before you sell this to anyone.
Affiliate link. We earn a commission if you subscribe, at no extra cost to you. 30 day free trial through HighLevel's Bootcamp offer, card required.
Against a human receptionist
| Option | Rough cost | The real trade |
|---|---|---|
| Voice AI | Cents per call | Answers instantly, every time, at 2am. Cannot handle anything genuinely unusual. |
| Answering service | Roughly $1 to $3 per call | A human, but a stranger with a script who does not know the business. |
| Part-time receptionist | Wages | Knows the business. Not there at 2am, and not there on Tuesday. |
| Voicemail | Free | Most callers do not leave one, and the ones who do have already called someone else. |
Set against voicemail, which is what most small businesses actually use after hours, Voice AI is not close. The comparison that matters is not AI against a person, it is AI against nobody answering at all.
For the text and chat side of AI Employee, see the Conversation AI guide.
Where it works, and where it does not
| Use case | Verdict | Why |
|---|---|---|
| After-hours reception for a trade or service business | Strong | The alternative is voicemail. Booking one extra job a month pays for it many times over. |
| Qualifying inbound leads from ads | Strong | Speed to lead is the whole game, and it answers on the first ring at 11pm. |
| Appointment reminders and confirmations | Strong | Scripted, predictable, and the calls are short so the per-minute cost barely registers. |
| Anything emotionally charged, complaints or cancellations | No | An upset customer meeting a bot is how you lose them for good. Route these to a person. |
| Complex technical or quoting conversations | No | It will confidently get something wrong, and a wrong quote on a recorded call is your problem. |
Worth watching the workshop before you build one
Voice AI is one of the few parts of this platform where a bad setup is worse than none. A badly scripted agent answering your main line loses real work, so it is worth seeing a proper build before you point it at live calls.
The Voice AI workshopWe earn a commission if you sign up, at no extra cost to you.
If voice AI is not actually why you are here
It is the most expensive thing to run on this platform and most businesses never switch it on. If what you need is a funnel, a list and a way to take payment, Systeme.io does that free with no usage meter attached.
Start free on Systeme.ioAffiliate link to Systeme.io. We earn a commission if you later upgrade, at no extra cost to you. No card required, and the free plan does not expire.
The version with no usage billing
ClickFunnels has no voice AI at all, which means no per-minute charges to forecast. A real limitation if you want an AI receptionist, and a real saving if you do not.
Start the ClickFunnels trialAffiliate link. We earn a commission if you subscribe, at no extra cost to you. 14 day trial, then $97 a month with a 30 day money-back guarantee.
Common questions
How much does GoHighLevel Voice AI cost per minute?
It depends which architecture you run. Speech-to-speech is a flat $0.10 a minute on Gemini 3.1 Flash Live Preview or $0.20 on GPT Realtime 2.1. The older cascading pipeline has no single rate and lands between about $0.060 and $0.215 a minute. Telephony is extra on both.
What is speech-to-speech Voice AI in GoHighLevel?
One model that hears and speaks directly, instead of three services passing text between them. It responds faster, switches language mid-sentence, and bills as a single flat per-minute rate rather than four separate meters.
Is GoHighLevel speech-to-speech Voice AI still free?
No. It was free until 6 September 2026, and agents left running on it started billing at the model rate automatically after that date. There was no prompt and no opt-in, so check any agent you set up during the free window.
Is speech-to-speech cheaper than the older Voice AI pipeline?
Only if you were using a good voice. At $0.10 a minute it is roughly double the $0.060 floor of the cheapest cascading setup and roughly half the $0.215 of the most natural one. The real gain is that $0.10 is the entire bill rather than the first of four lines.
Does AI Employee Unlimited cover Voice AI calls?
It covers the AI, not the phone call. Telephony bills separately at Twilio rates on every plan, so an inbound Voice AI call always costs you twice.
Is GoHighLevel Voice AI cheaper than an answering service?
Substantially. Cents per call against roughly a dollar to three per call for a human service. But the fair comparison for most small businesses is against voicemail, which is what actually happens to their after-hours calls now.
Which Voice AI voice should I choose?
The cheapest option unless the voice quality genuinely matters to the brand. Moving to the most natural voice takes your floor from about $0.060 to about $0.215 a minute, nearly four times, for a difference many callers will not consciously notice on a 90 second booking call.
When should Voice AI not answer?
Complaints, cancellations and anything emotionally charged, plus complex quoting. An upset customer meeting a bot is how you lose them, and a confidently wrong quote on a recorded call becomes your problem. Route those to a person.
Related: all GoHighLevel AI costs and the full pricing breakdown.
Last reviewed September 3, 2026