5 min read
$4.80 an hour. That is what xAI’s new Grok Voice 3.0 model costs to run a live phone conversation through its developer API, at a reported base rate of $0.08 per minute, the same price as the version it replaced on October 14. For a solo owner who misses calls while on a ladder, in a session, or behind the wheel, that number sounds like the end of voicemail.
It is not quite that. A cheap voice model is one ingredient, not a receptionist. This piece separates what the falling price actually changes for a one-person business from what it does not, and gives you a plan for the next six months.
What the price drop really means
A “voice agent” is software that listens to a caller, works out what they want using an AI model, and speaks back in a natural voice, all in real time. Two years ago, running one cost enough that only call centers bothered. Now several providers sell the core voice layer for a few cents a minute. xAI publishes its model lineup and pricing on its developer documentation, and coverage of the October release notes that the price did not move between versions while the model improved.
Run the arithmetic for a typical solo service business. Say you get 120 calls a month averaging three minutes. That is 360 minutes, or roughly $29 a month in raw voice cost at $0.08 a minute. That figure is real, but it is also the most misleading number in this whole category.
Why $0.08 a minute is not your real cost
The per-minute rate is the price of the engine. To actually answer your phone, you also need:
- A phone number and call routing, usually through a telephony provider, billed separately per minute and per number.
- Something that knows your business: your hours, prices, service area, and what to do with an angry caller. Someone has to write and maintain that.
- A way to act, such as booking into your calendar or texting you a summary. That is integration work.
- Testing time. Every voice agent mishears names, numbers, and addresses some of the time. You need to listen to recordings and fix what breaks.
The developer API is built for people who can wire those pieces together. Most solo owners will not touch it directly. They will buy a packaged AI receptionist app that uses a model like this underneath and charges a monthly fee on top. The real story is that those packaged products should get cheaper and better over the next year, because their biggest cost line just stopped going up.
The skeptic’s checklist before you switch anything on
Vendor benchmarks for voice models, including xAI’s, are the companies’ own numbers, and independent tests lag behind. So judge any product on your own calls, not on a launch post. Four questions worth asking:
- Does it say it is an AI? Callers forgive a bot that admits it is one. They do not forgive one that pretends. We made the same argument about video in AI video agents can now pass as human.
- What happens when it is unsure? The right answer is “takes a message and texts you”, not “makes something up about your prices”.
- Can you hear every call? You want recordings or transcripts for at least the first month.
- What are the rules where you work? Some US states now have AI disclosure and call recording rules. Our plain-English guide to state AI laws is a starting point.
Inbound is fine. Outbound is a different animal
This is where solo owners can get into real trouble. Answering calls that customers make to you is low risk. Using an AI voice to call people, for reminders, follow-ups, or sales, is not. In 2024 the US Federal Communications Commission ruled that AI-generated voices count as “artificial” voices under the Telephone Consumer Protection Act (TCPA), the federal law on robocalls. That means the consent rules for prerecorded calls apply.
Plain-English takeaway: do not let any AI voice tool place calls to customers unless you have their clear written consent for that kind of call, and talk to someone qualified before you run anything that looks like a calling campaign. Texts you send yourself are a far safer reminder channel.
A six-month plan for a one-person business
Here is how we would approach it, in order:
- Month one: count your missed calls. Most phone carriers and business phone apps show missed call logs. If you miss fewer than ten a month, a good voicemail greeting and a fast text-back habit beat any AI.
- Month two: write your “front desk sheet”. One page with hours, service area, three most common questions with answers, prices you are happy to quote, and what counts as urgent. You need this whatever tool you pick.
- Month three: trial one packaged receptionist on after-hours calls only. Keep a human (you) answering during the day. Compare.
- Months four and five: review every transcript weekly. Fix the sheet where the agent stumbled. Track how many calls turned into bookings.
- Month six: decide on the numbers. Keep it if it books more than it costs and nobody complained. Cancel if not.
A narrow job works best. The waitlist example in our dog groomer case study shows how one tightly scoped AI task can pay off without trying to replace the whole front desk.
The bottom line on cheap AI voices
Falling voice prices are genuinely good news, but mostly for the companies that build receptionist products, and only later for you. The smart move for a solo business is not to rush into the cheapest API. It is to get your front desk information written down now, so that when a well-reviewed, affordable product appears, you can switch it on in an afternoon and trust what it says.
How many calls did you miss last month? If you do not know, that is the first number to find.
