Voice AI Cost: How to Calculate Your Monthly Spend

Voice AI cost for a month is the number of minutes your agent spends on calls multiplied by its per-minute rate, plus telephony after transfers, plus fixed monthly items such as phone numbers and extra concurrency. Getting a reliable estimate depends less on the vendor's headline rate and more on knowing your own call volume, how long the agent stays on each call, and how often it hands off to a person.

This guide walks through the inputs, the two formulas, and a full worked example using Retell AI's published component rates as of October 2026.

What goes into the monthly cost of a voice AI agent?

Four buckets cover almost every invoice:

• Per-minute usage: the conversation itself. On Retell this is itemized as voice infrastructure, the language model, and text-to-speech.
• Telephony: the phone line. It runs for the whole call, including any time after a transfer.
• Per-minute or per-event add-ons: features such as a knowledge base, denoising, PII removal, batch dialing or branded caller ID.
• Fixed monthly items: phone numbers, concurrency above the included amount, extra knowledge bases and verified numbers.

Some vendors also charge a platform fee or require an annual minimum. Retell's pay-as-you-go plan has neither, but if you are comparing quotes, the voice AI pricing comparison explains how those structures differ.

What inputs do you need before you calculate?

Pull these from your current phone system or contact center reports before opening any calculator.

• Monthly call volume the agent will answer or place
• Average and median minutes the agent will be on each call
• Share of calls transferred to a person, and how long they last after transfer
• Calls in your busiest hour, to size concurrency
• The LLM and voice you plan to use
• Add-ons you need: knowledge base, denoising, PII removal, batch dialing
• Phone numbers: bought through the platform or brought over SIP

Use the agent's expected time on the call, not your human average handle time. Human AHT includes hold and after-call work that an AI agent does not have, and it ignores the calls the agent will transfer early. Check the median duration as well as the average. A few very long calls can pull the average up and overstate spend.

How do you calculate voice AI cost per minute?

Add up each component the platform bills per minute. On Retell the core components, from the Retell pricing page, are:

• Voice infrastructure: $0.055 per minute, the same for every agent.
• LLM: varies by model. For example, GPT 5 mini is $0.008, GPT 5.1 is $0.04, GPT 5.4 is $0.08 and GPT 5.5 is $0.16 per minute on the standard tier. Fast tiers cost more.
• Text-to-speech: $0.015 per minute for Retell platform voices and several other providers, up to $0.10 per minute for ElevenLabs v3 or v4.
• Telephony: $0.015 per minute on a Retell US number. There is no Retell charge when you bring your own carrier over SIP, as described in the guide to keeping your existing phone number with a SIP connection. Your carrier still bills you separately.

Per-minute rate formula: Per-minute rate = Voice infrastructure + LLM + Text-to-speech + Telephony + Per-minute add-ons

$0.055 voice infra + $0.04 GPT 5.1 + $0.015 platform voice + $0.015 telephony + $0.005 knowledge base = $0.13 per minute on Retell pay-as-you-go rates

Price the exact model and voice you will run, since the LLM is the component that varies most.

Retell advertises $0.07 to $0.31 per minute for voice agents. Its docs describe that range as voice infrastructure, text-to-speech and the LLM together, depending on the model and voice, so telephony and add-ons sit on top. The range also does not cover every option on the same page. GPT 5.5 on its fast tier is $0.32 per minute for the LLM alone, which comes to $0.39 per minute with voice infrastructure and a $0.015 voice, before telephony. Price the exact model and voice you plan to run.

How do you calculate the monthly total?

Multiply AI minutes by the per-minute rate, add telephony for the time callers spend with your team after a transfer, then add fixed monthly items.

Monthly cost formula: Monthly cost = (AI minutes × per-minute rate) + (post-transfer minutes × telephony rate) + fixed monthly items

(150,000 × $0.13) + (60,000 × $0.015) + $20 numbers + $160 concurrency = $19,500 + $900 + $180 = $20,580 per month

AI minutes are calls multiplied by the agent's time on each call, not your human average handle time.

Here is a full example for an inbound support line handling 50,000 calls a month. The agent spends an average of 3 minutes on each call and transfers 20% of callers, who then spend another 6 minutes on the line with a person. The agent uses GPT 5.1, Retell platform voices, a knowledge base and Retell telephony, and the team plans for 40 simultaneous calls at peak.

• AI minutes: Calculation: 50,000 calls × 3 min × $0.13. Monthly cost: $19,500.
• Telephony after transfer: Calculation: 10,000 calls × 6 min × $0.015. Monthly cost: $900.
• Phone numbers: Calculation: 10 × $2.00. Monthly cost: $20.
• Extra concurrency: Calculation: (40 − 20 free) × $8.00. Monthly cost: $160.
• Total: Calculation: . Monthly cost: $20,580.

That comes to $20,580 a month, or about $0.41 per call. Two changes show how sensitive the estimate is. Switching the LLM to GPT 5.5 raises the per-minute rate from $0.13 to $0.25 and the monthly total to $38,580. Moving telephony to your own SIP carrier removes the $0.015 Retell telephony charge and the Retell number fees, but your carrier bill replaces them.

How do transfers, voicemail and silence change the bill?

Billing rules decide which seconds count. On Retell, per the pricing page and billing docs:

• Per-second billing: each call is tracked to the nearest second and billed on total accumulated minutes, with no rounding up per call.
• Transfers: the AI agent fee stops once the caller is connected to the destination and the agent drops off. Telephony continues for the rest of the call. During the hold and briefing phases of a warm transfer, the agent is still on the line and bills at the normal rate.
• Silence and hold: billed, because speech recognition keeps listening for the whole call.
• Failed calls and voicemail: calls that never connect are not billed. Calls that reach voicemail bill only while the agent is active.

Transfers are where most estimates go wrong. A transferred call still costs its AI minutes plus a person's time. To see how that affects the cost of a solved issue, read the cost per resolution benchmarks.

Which fixed monthly charges should you add?

Retell lists these subscription items separately from per-minute usage:

• Retell phone numbers: $2.00 per number per month. No charge for numbers you bring through custom telephony.
• Concurrency: the first 20 simultaneous calls are free, then $8.00 per extra concurrent call per month. Enterprise plans have custom concurrency starting at 50+ calls.
• Knowledge bases: the first 10 are free, then $8.00 per knowledge base per month.
• Verified phone numbers: $10.00 per number per month.
• Retell SMS: $20.00 per month, with no charge for SMS on custom telephony.

To size concurrency, estimate peak simultaneous calls as busiest-hour calls multiplied by average minutes, divided by 60, then add headroom. 600 calls in the busiest hour at 3 minutes each is 30 simultaneous calls on average. The Settings → Limits page in the Retell dashboard has a calculator that recommends a value.

What else should the budget cover?

A few charges only show up once you are live or testing heavily:

• Testing: test calls bill at production rates. Text simulations bill per message, and web and phone test calls bill per minute.
• AI quality assurance: the first 100 analyzed minutes per workspace are free, then $0.10 per analyzed minute.
• Long prompts: for agents whose prompts exceed 4,000 tokens, Retell scales billed duration by tokens divided by 4,000. A 60-second call at 4,800 tokens bills as 72 seconds.
• Concurrency burst: calls above your concurrency limit can proceed if burst is enabled, with a $0.10 per minute surcharge on the whole call.
• Outbound extras: batch calling adds $0.005 per dial and branded caller ID adds $0.10 per outbound call.

Outbound campaigns need one more input: the share of dials that connect. Dials that never connect are not billed for minutes, but the batch calling fee applies per dial. As an illustration, 20,000 dials a month with 30% connecting and 2 minutes per connected call is 12,000 minutes. At $0.13 per minute that is $1,560, plus $100 in batch dial fees. Branded caller ID adds $0.10 per outbound call, so on a high-volume campaign it can cost more than the minutes themselves. Check it against the answer-rate lift it buys before turning it on for every list.

Enterprise pricing on Retell is custom, so high-volume teams should treat pay-as-you-go rates as the upper bound for planning.

How do you check the estimate after launch?

Compare the forecast against real usage in the first two to four weeks. In Retell, the Usage tab on the Billing page shows total cost, call minutes, average cost per minute, and cost split across voice infrastructure, LLM, telephony and concurrency. If the average cost per minute is above your forecast, check the model and voice first, then prompt length. If total minutes are above forecast, look at average call duration by agent version. A new prompt that adds a confirmation step to every call adds minutes to every call.

Re-run the estimate whenever volume or the agent changes: a new use case, a seasonal peak, a different model or a longer prompt. Keeping the inputs in one sheet makes that a quick update instead of a new project.

To keep spend inside a ceiling, workspaces on credit-based billing can set auto recharge thresholds and, where available, a monthly budget that pauses calls at 100%. Track both the bill and the outcomes it buys, since usage-based spend rises with volume. The explainer on how usage-based pricing works covers the billing mechanics in more depth.

The fastest way to replace assumptions with real numbers is to build one agent, place test calls, and read the Usage tab. You can start free in the Retell dashboard with $10 in credits.