Vapi Pricing for UAE Businesses: Per-Minute Costs Explained

Almost every conversation we have about Vapi pricing starts the same way: someone has read the pricing page, seen $0.05 per minute, and built a budget around it. The number is accurate. The budget is usually wrong by a factor of six or seven, because $0.05 buys one layer of a five-layer stack.

This post rebuilds the math from published rate cards only. Every figure below is quoted from an official pricing page and linked in the sources at the end. Where a vendor does not publish a per-minute number, we say so instead of inventing one.

What does Vapi actually charge?

Vapi sells two tiers. Build is the usage-based tier that nearly every UAE business starts on. Scale is an annual contract with a fixed platform fee and committed volume, priced on request. The published Build rate card is short:

The critical phrase on that page is "Excludes Model Provider Costs". Vapi is not marking up the models. It passes them through at cost, and charges you $0.05 a minute to orchestrate them. That is a genuinely fair way to price a platform. It is also why the $0.05 tells you almost nothing about your bill.

What does a UAE minute actually cost?

Here is where the UAE changes the arithmetic. A voice agent minute is five costs stacked on top of each other: telephony carries the call, STT hears it, an LLM decides what to say, TTS says it, and Vapi orchestrates the whole pipeline. Four of those five are not Vapi's.

Telephony is the one that hurts. Twilio's published UAE voice rate card lists $0.2995 per minute for UAE mobile and $0.3635 per minute for UAE local numbers. Those are not typos, and they are not unusual for this market. UAE termination is expensive compared with the US or UK rates most Vapi tutorials quietly assume.

Read that bottom row carefully. Before a single token of LLM inference or a single character of synthesised speech, one minute of an inbound UAE mobile call costs $0.3553. Vapi's share of that is $0.05, or roughly 14%. Telephony is roughly 84%. If the caller dials a UAE landline instead of a mobile, the floor rises to about $0.4193 per minute on the same published rates.

This is the single most useful thing to understand about Vapi pricing in this market: the platform fee is not the variable worth optimising. Halving Vapi's fee would save you $0.025 a minute. Getting your telephony right is worth ten times that.

So what is the price per month?

"Per minute" is how the vendors price. "Per month" is how businesses budget. Converting between them needs one honest input: your actual answered-minutes volume. Using the same published floor of $0.3553 per UAE mobile minute:

• 500 answered minutes/month (a small clinic or single-branch business): about $177.65, plus roughly $1/month for the number. Vapi's own share of that is $25.
• 1,000 answered minutes/month: about $355.30. Vapi's share is $50.
• Concurrency: 10 simultaneous lines are included. An eleventh line costs $10/month. Most SMEs never touch this. A campaign dialling 40 numbers at once does, immediately.

Why can't anyone quote LLM and TTS per minute?

Because neither is billed by time. This is where most cost calculators start guessing, so it is worth being precise about the mechanism rather than pretending there is a number.

LLMs bill per token. A minute of conversation might be two short turns or eight long ones. It might carry a 200-word system prompt or a 2,000-word one, re-sent on every turn. A retrieval step that injects your product catalogue into context can multiply the token count of an otherwise identical minute. Two businesses on the same model, same platform and same call length can differ several-fold. Anyone quoting you a flat LLM cost per minute has picked an assumption and hidden it.

TTS bills per character. ElevenLabs publishes plan tiers rather than a per-minute voice rate: Starter at $6/month for 30k credits, Creator at $11/month for 121k, Pro at $99/month for 600k, Scale at $299/month for 1.8M, Business at $990/month for 6M. On the V2 multilingual models, one text character equals one credit; optimised models run between 0.5 and 1 credit per character. So your TTS cost tracks how much your agent talks, not how long the call lasts. A concise agent is measurably cheaper than a chatty one, which is a design decision, not a billing one.

STT is the one model layer that genuinely does bill per minute, which is why it appears in the table above. Deepgram publishes $0.0048/min for Nova-3 Monolingual streaming and $0.0058/min for Nova-3 Multilingual. That multilingual premium of one-tenth of a cent is not worth a second thought in Dubai, where a single afternoon brings calls in English, Arabic, Hindi and Malayalam. Take the multilingual model.

Why is telephony the biggest line item here?

The $0.2995 to $0.3635 per minute on Twilio's UAE card reflects the cost of terminating a call on a UAE network through an international carrier. That is the price of convenience: you can provision globally and be live quickly, and you pay international termination on every minute of every call, forever.

The alternative is local SIP through a UAE-licensed operator, terminating on local infrastructure. The licensed operators do not publish a public per-minute rate card, so we will not invent one for them. What we can say from doing the work is that this route trades a higher setup cost and real integration effort for a materially lower cost per minute, and it becomes the obvious choice as volume rises. Our breakdown of the hidden costs in cheap voice agent quotes covers what that integration actually involves.

We build on this route in production. The IT World Trading LLC deployment runs a multilingual outbound agent on TDRA-compliant telephony, installed on-site with local SIP hardware rather than an international CPaaS hop. The setup work is not trivial, and it is exactly the work that the $0.05-per-minute mental model leaves out entirely.

Does the math favour building on Vapi yourself?

Sometimes, genuinely. Vapi is a well-built platform priced honestly, and if you have an engineer who enjoys this work, the per-minute economics are excellent. But the per-minute cost is not the whole cost, and the $0.05 figure is seductive precisely because it makes the rest invisible.

What the rate card does not price: prompt engineering and the iteration cycle that follows, telephony provisioning and carrier configuration, CRM and calendar integration, TDRA compliance configuration for any outbound activity, voice selection and language QA across each language you serve, and the ongoing monitoring that catches an agent quietly failing in Arabic while passing in English. Those are the line items that decide whether the deployment works, and none of them appear on any pricing page.

The honest framing is that Vapi's pricing is an infrastructure cost, not a solution cost. If you want the full comparison of the platform-versus-done-for-you decision, we have written it up in detail: Vapi and Retell versus done-for-you deployments, and every price factor in a Dubai AI receptionist quote.

If you would rather not assemble the stack yourself, that is what we do. See our AI receptionist deployments in Dubai, AI appointment booking, or the full service list.

What is the bottom line on Vapi pricing in the UAE?

Vapi charges $0.05 a minute and passes models through at cost. That is a clean, fair rate card. In the UAE, it is also about 14% of your true per-minute floor, because published UAE termination runs $0.2995 to $0.3635 a minute and dominates everything else on the invoice.

Budget from the floor, not the headline. Ask any vendor quoting you a per-minute price which telephony route they are using, and whether LLM and TTS are inside or outside the number. If they cannot answer both questions precisely, the quote is a guess.