What a GoHighLevel Voice AI Minute Actually Costs (2026)

Voice AI billing generates more confusion, and more billing complaints, than any other GoHighLevel feature. The reason is structural: a single AI phone call runs four meters at once, three of them under the AI umbrella and a fourth on the phone system. Users who expect one per-minute price discover their wallet draining from charges they never mapped to line items.

This breakdown prices each meter with the current official rates, runs worked examples at realistic volumes, and shows exactly where the breakeven sits between pay-per-use and the flat AI Employee plans. Every rate below comes from HighLevel's official AI pricing documentation, verified July 16, 2026. HighLevel adjusts rates periodically (the current voice engine rate took effect May 20, 2026), so treat the numbers as current, not permanent.

The three AI meters on every Voice AI call

Meter 1: the voice engine, $0.045 per minute

The base engine fee is the simplest meter: call duration in minutes multiplied by $0.045. A 10 minute call is $0.45. This rate applies regardless of which voice or model is selected and covers the real-time orchestration that keeps latency low enough for natural conversation.

Meter 2: text to speech, the biggest variable

The voice the agent speaks with is billed per minute on top of the engine fee, and the spread is wide:

The gap deserves emphasis. ElevenLabs V3 costs more than eleven times the OpenAI and Cartesia voices, and on a per-call basis the premium voice costs nearly three times the engine fee itself. Combined engine plus TTS lands at $0.06 per minute on the budget stack and $0.215 per minute on the premium one. Nothing else in the Voice AI bill moves the total like the voice dropdown does.

Meter 3: LLM tokens, the meter that sounds scary but is not

The model powering the conversation bills by tokens, at rates per million that range from $0.05 (GPT-5 Nano) through $0.10 (Gemini 2.5 Flash Lite), $0.25 (GPT-5 Mini), $1.25 (GPT-5), and up to $3.00 for Claude 4.5 Sonnet, with output tokens priced higher than input on the frontier models.

In practice this is the smallest meter by far. Spoken conversation generates modest token volume: people speak far fewer words per minute than an LLM can read. As an illustration of scale, even if a call somehow consumed 100,000 tokens, that would cost three cents at GPT-5 Mini input rates. Real calls consume far less. Model choice matters for answer quality and instruction following; it rarely matters for the bill.

The fourth meter: telephony

Phone system charges apply to every call, on every plan, including AI Employee Unlimited. The AI meters above are charged for the intelligence; the carrier minutes, number rental, and any A2P registration are charged by the phone system separately. Anyone auditing a surprising "Voice AI bill" should split it in two, because the telephony half often drives more of the total than the AI half. The phone system guide covers those rates and the A2P 10DLC setup in detail.

Worked examples at real volumes

HighLevel's own reference example: a 10 minute call on an OpenAI voice costs $0.45 engine plus $0.15 TTS, $0.60 total before tokens and telephony. Scaling that math to monthly volumes, always on pay-per-use rates and before telephony:

• A dental office capturing after-hours calls. Around 60 calls a month averaging 4 minutes: 240 minutes. Budget stack: about $14.40 plus token cents. Premium ElevenLabs V3 voice: about $51.60. The voice choice alone turns a rounding error into a real line item.
• An agency running missed-call answering for one client. Around 300 calls averaging 3.5 minutes: 1,050 minutes. Budget stack: about $63. Premium stack: about $226. At this volume the budget stack still undercuts the $97 Unlimited plan, but the premium stack blew past it long ago.
• A front desk replacement handling most inbound. 2,000 minutes and up. Budget stack: $120 and climbing, so Unlimited at $97 flat is cheaper before the quality-of-life benefit of a predictable invoice.

The breakeven arithmetic is simple enough to run per sub-account: at $0.06 per minute the pay-per-use bill crosses $97 at roughly 1,600 minutes per month, which is under an hour of AI calls per day. On an ElevenLabs V3 voice at $0.215 per minute, breakeven collapses to roughly 450 minutes per month, barely 15 minutes a day. Voice-heavy accounts on premium voices should almost never be on pay-per-use.

Where the flat plans fit

The plan mechanics, covered plan by plan in the full GoHighLevel AI pricing breakdown, interact with Voice AI like this:

• Pay-per-use (no monthly fee): every meter runs at the rates above. Right for pilots, seasonal use, and accounts under a few hundred minutes a month.
• AI Employee Growth, $50 per month per sub-account: includes 100 Voice AI minutes. Past 100 minutes, overages bill automatically at pay-per-use rates without interrupting calls, so a busy month can quietly cost well over $50. Growth suits chat-first accounts that dabble in voice, not voice-led accounts.
• AI Employee Unlimited, $97 per month per sub-account: unlimited Voice AI minutes, with only telephony still metered. The moment Voice AI is the product a sub-account sells, this is the tier.

For agencies packaging AI receptionists, the margin play is rebilling: HighLevel lets agencies mark up AI usage or resell AI Employee plans at their own prices, but platform rebilling of AI requires the $497 Agency Pro (previously SaaS Pro) plan. An agency charging a client $297 for an AI receptionist that costs $97 plus telephony is a common arrangement, and it only works cleanly on Agency Pro.

How the charges actually hit the account

Understanding the meters is half the picture; knowing where the money leaves the account is the other half. Voice AI usage charges draw from the sub-account wallet, the same prepaid balance that funds SMS, email, and phone charges. When the wallet hits its recharge threshold, it tops up automatically from the card on file. This design is why Voice AI surprises feel sudden: nothing interrupts service as spending climbs, the wallet just recharges more often, and the damage only becomes obvious on the card statement or in the billing report.

Three habits keep it visible. First, check the wallet transaction log weekly during the first month of any Voice AI rollout, because that is where engine, TTS, and telephony charges appear as separate entries and where the real per-call cost reveals itself. Second, set the auto-recharge amount deliberately rather than leaving defaults, so a runaway week shows up as frequent recharge notifications instead of one large silent charge. Third, agencies managing multiple sub-accounts should review usage per location, since one client's call spike can hide inside an agency-level total.

For agencies that rebill, the wallet is also the audit trail: the platform's usage reports per sub-account are what justify the marked-up invoice to the client, so clean per-location tracking is a revenue practice, not just a cost control.

Five levers that actually cut the bill

1. Audit the voice selection first. Downgrading ElevenLabs V3 to V2.5 cuts TTS cost by about 80 percent; moving to Cartesia or OpenAI cuts it by about 91 percent. Test whether callers can even tell.
2. Cap call duration in the agent settings. The engine and TTS meters bill by the minute, so a rambling agent is an expensive agent. Tight prompts that resolve or route calls quickly are a direct cost control.
3. Transfer to humans at the right moment. Handing a qualified caller to staff ends the AI meters. GoHighLevel's call transfer options (web, phone, and SIP) hand the caller over without a callback, and a well-placed transfer beats a 10 minute AI conversation on both cost and conversion.
4. Pick the model for the job, not the badge. The token meter is small, but there is no reason to pay Claude 4.5 Sonnet rates for appointment booking that GPT-5 Mini or Gemini 2.5 Flash Lite handles reliably. The Voice AI setup and testing guide covers model behavior differences in practice.
5. Re-run the plan math quarterly. Usage drifts, and both directions cost money: pay-per-use accounts that grew past breakeven, and Unlimited accounts whose usage collapsed after a season.

Summer of AI and the July Voice AI push

Voice AI is GoHighLevel's featured theme for July 2026 under the Summer of AI promotion, which runs through August 31. The promotion offers 30 day AI Agent trials, extendable to 60 days with an AI Employee plan, which is a legitimate way to measure a sub-account's real minute volume before choosing a tier.