Pay for what you use. About 5¢ a minute, all in.
A voice agent costs 4.2¢ a minute for speech recognition, the voice and everything around them, plus the language model by the token: about 5.2¢ for a typical minute of conversation. Prepaid, billed by the second, no plan and no monthly fee.
4.2¢ for the platform plus about 1¢ of language model, on the web or in your app.
A call your agent answers on a Spanish number: the agent minute plus 3.5¢ of line. Every country below.
The voice your agents use, as an API: streaming or one request.
Transcription on its own. Noise removal costs the same.
Prices in US dollars, before tax. The language model is priced at a typical prompt (about 20,700 input and 1,300 output tokens a minute); a longer prompt costs more. Dedicated instances, on-prem and on-device licensing are quoted separately.
Every line on the bill.
| Line | Price | How it’s measured |
|---|---|---|
| Voice agent, platform | $0.042 a minute | Per second of session. Covers speech recognition, the voice, turn-taking, noise removal, orchestration and call recording. |
| Language model, input | $0.44 per million tokens | Every request the agent makes to the model: replies, rounds after a tool, greetings, summaries, the post-call evaluation. |
| Language model, output | $0.94 per million tokens | Not billed for an agent on your own model key. |
| Phone line | By country and direction | Per started minute, on top of the agent. Table below. |
| Phone number | $5 a month | From the balance. Local numbers in the US, Spain, the UK, Germany, France, Italy and Portugal. |
| Call logged in your CRM | $0.04 a call | Only for agents with your CRM turned on. |
| Web search | $0.15 a search | Only for agents with web search turned on; a search that fails is free. |
| Text-to-speech API | $4.10 per million characters | Per character of text synthesized. |
| Speech-to-text API | $0.0014 a minute | Per second of audio. |
| Noise removal API | $0.0014 a minute | Per second of audio. |
Phone lines, per minute
| Country | Call received | Call made to a landline | Call made to a mobile |
|---|---|---|---|
| Spain | $0.035 | $0.049 | $0.105 |
| Austria | $0.028 | $0.027 | $0.087 |
| Belgium | $0.035 | $0.084 | Not available |
| Canada | $0.021 | $0.025 | Landline rate |
| France | $0.028 | $0.025 | $0.084 |
| Germany | $0.028 | $0.028 | $0.093 |
| Ireland | $0.025 | $0.041 | $0.077 |
| Italy | $0.035 | $0.047 | Not available |
| Luxembourg | $0.025 | $0.084 | Not available |
| Netherlands | $0.035 | $0.042 | Not available |
| Portugal | $0.025 | $0.066 | Not available |
| Switzerland | $0.028 | $0.098 | Not available |
| United Kingdom | $0.025 | $0.022 | $0.077 |
| United States | $0.019 | $0.025 | Landline rate |
On top of the agent, per started minute: a 61-second call is two minutes of line. A received call is priced by the country of your number, a call your agent makes by the destination. A call nobody answers costs nothing. Mobiles in Belgium, Italy, Luxembourg, Netherlands, Portugal and Switzerland can’t be called: carrier rates there are above our ceiling. Calls your agent makes and phone numbers need a first top-up. The same card, as JSON: api.lokutor.com/v1/pricing (version 2026-10-09.1).
What things cost.
A 3-minute call to your agent on a Spanish number
8.7¢ a minute. At 3 minutes 10 seconds the line is 4 started minutes, 3.5¢ more.
1,000 minutes of conversations on the web
With speculative replies off, $48.79.
A receptionist: 300 minutes of calls a month on a Spanish number
Calls received, agent and line included.
1 million characters through the text-to-speech API
Transcribing an hour of audio costs 8.4¢.
At a typical prompt, before tax. What each of your calls really cost is itemised in the dashboard.
Faster replies, if you want them.
Every agent starts with speculative replies on: it begins its answer while the caller pauses, so when they have finished, the reply is already on its way. When they keep talking, that early reply is thrown away and its tokens are on your bill. Turn it off per agent if you’d rather not pay for replies nobody hears.
| On (default) | Off | |
|---|---|---|
| A typical web minute | 5.2¢ | 4.9¢ |
| Reply time, on turns it catches | about 0.3 s | about 0.66 s |
| Reply time, on average | about a quarter of a second faster with it on | |
How you’re billed. All of it.
- Top up$10 to $2,000 a payment, from the dashboard. You pay in your own currency at checkout; the balance is kept in US dollars. For more than $2,000 at once, write to us.
- TaxesPrices are before tax. VAT or sales tax is added at checkout: a $10 top-up by an individual in Spain costs $12.10 and adds $10. A business can enter its VAT ID; EU businesses outside Spain are reverse-charged. Ask for an invoice when you top up.
- Welcome credit$5 when you confirm your email, valid 90 days: about 95 minutes of conversation. It doesn’t pay for calls to phones or for phone numbers; those need a first top-up.
- Auto-rechargeOptional. By default, when the balance falls below $5 we add $20, up to $500 a month. Charged in US dollars to your saved card.
- MeasuredAgent time and audio by the second, the language model by the token, phone lines by the started minute, as carriers bill them. No minimums.
- ItemisedEvery call in the dashboard shows what it cost: platform, model tokens by kind, phone line and extras.
- Calls at once3 until your first top-up, 10 after, more on request.
- Running lowA call in progress is never cut mid-sentence: it can take the balance to −$2, then the agent says goodbye and hangs up. Starting a call or a request needs $0.25 of balance.
- Expiry and refundsPurchased balance never expires. Ask within 14 days of a top-up and we refund what you haven’t used. Credits we give expire and aren’t refunded.
- Price changesEach call is charged at the prices in force when it started. Increases are announced 30 days ahead; decreases apply at once.
What would it cost you?
Set your usage. The maths is shown, with the prices from the live rate card.
- Platform, 1,000 min × 4.2¢$42.00
- Language model, typical prompt$10.33
- No phone calls$0.00
- No phone numbers$0.00
Prices from the rate card, version 2026-10-09.1.
The language model is priced at a typical prompt; a longer prompt uses more tokens. Phone lines are billed per started minute, so the estimate adds half a minute to each call. Your real usage is itemised call by call in the dashboard.
The whole minute, next to what others publish.
| Platform | Published price per minute | What that covers |
|---|---|---|
| Lokutor | about 5.2¢ | 4.2¢ platform with recognition and voice, plus the language model by the token (about 1¢ at a typical prompt) |
| Cartesia | 6¢ | Call duration |
| Deepgram Voice Agent | 7.5¢ | Standard tier, pay as you go |
| ElevenLabs Agents | 8¢ | Per minute beyond a plan’s allocation |
| Retell AI | 5.5¢ + LLM + voice | Voice engine; language model and TTS billed on top |
| Vapi | 5¢ + providers | Platform fee; recognition, LLM and TTS billed on top |
Each company’s own pricing page, last checked 5 October 2026 — follow the links to check. Lokutor at a typical prompt. Web calls; telephony is extra on every platform, ours included. Full comparisons
Common questions.
How much does a voice AI agent cost on Lokutor?
About 5.2¢ a minute on the web or in an app at a typical prompt: 4.2¢ for the platform (speech recognition, the voice, turn-taking, noise removal, orchestration and recording) plus the language model by the token, about 1¢. A phone call adds the line: a call answered on a Spanish number comes to about 8.7¢ a minute. Published rates elsewhere are 6–8¢ a minute, and several platforms bill the language model and the voice on top.
Do I need a plan or a subscription?
No. There is no plan and no monthly fee: you top up a balance from $10 and it is spent as you use it. Phone numbers are the only monthly charge, $5 each, taken from the balance.
Can I try it for free?
Yes. New accounts get $5 of credit when they confirm their email, about 95 minutes of conversation, valid for 90 days. No card needed. Calls to phones and phone numbers need a first top-up.
Why does the language model cost extra?
Because it depends on you. Every turn sends the model your agent’s instructions and the conversation so far, so a long prompt costs more than a short one. We bill the tokens each call really used, at one price whichever model provider answers, and show them call by call. An agent on your own model key pays no tokens to us.
What are speculative replies?
The agent starts answering while the caller pauses, so turns where the caller has finished get a reply in about 0.3 seconds instead of about 0.66. When the caller keeps talking, the early reply is thrown away, and its tokens are billed: about 0.35¢ a minute at a typical prompt. It is on by default, and you can turn it off per agent.
Are prices with or without VAT?
Without. VAT or sales tax is calculated at checkout from your country and added to the top-up; the balance gets the amount before tax. A business can enter its VAT ID: an EU business outside Spain is reverse-charged. Ask for an invoice when you top up.
Does my balance expire? Can I get a refund?
Balance you buy never expires. Ask within 14 days of a top-up and we refund the part you haven’t used, in the currency you paid. Credits we give, like the welcome credit or a promotional code, expire and aren’t refundable.
What happens when my balance runs out?
A call in progress is never cut mid-sentence. It can take the balance down to −$2; then the agent apologises, says goodbye and hangs up, without telling the caller why. New calls and API requests need at least $0.25 of balance, and calls to your number aren’t answered until you top up. Auto-recharge tops you up before that happens, if you turn it on.
In which currency do I pay?
The balance is in US dollars. At checkout you pay in your own currency, converted by Stripe, or in dollars if you prefer. Auto-recharge is charged in dollars and converted by your bank.
I’m on a monthly plan. What changes?
Plans are being replaced by the balance. Every account registered before the change gets a credit when it moves over.
Why is it cheaper?
Our voice model is a few tens of millions of parameters, not billions, and runs on commodity hardware — a fraction of what the large GPU models behind most voice platforms cost to serve. That is what lets us price the platform at 4.2¢ a minute with recognition and the voice inside it, and run a sustainable business on it. It is not a promotional price.
Which languages are supported?
Spanish, Catalan, Galician, Basque, Portuguese, English, French, Italian and German — including the Iberian languages most voice platforms don’t offer at all.
Can voice AI run on-premise?
Yes. Lokutor's full pipeline — turn-taking, noise suppression, speech-to-text and speech synthesis — runs inside your own infrastructure with audio that never has to leave your environment. Ask through the Enterprise page.
What is on-device or edge voice AI?
Running the voice pipeline directly on the compute already inside a product — a device, robot or appliance — instead of round-tripping to the cloud. Lokutor's models are small enough to run in real time on an ordinary CPU, so they fit on-device without adding a per-unit cloud inference bill. See On-device for hardware makers.
Start with $5 of credit.
About 95 minutes of conversation, no card. Then about 5¢ a minute, all in.
More calls at once, your own servers or your device: Enterprise.