Voice AI cost calculator
Pick the platform you run on, enter your minutes a month and what you pay on top of its rate, and see an estimated monthly cost and cost per connected minute. Next to it, the same minutes priced on the paths from our worked example.
Estimate from list prices, not a client result. Platform rates and plan fees are the LIST prices on our voice AI price table, checked 24 October 2026. The worked example's paths add its LIST server prices, checked 21 October 2026, and its ESTIMATE of an LLM line. Deepgram's Growth commitment is spread over 12 months as an ESTIMATE. Not a contract price, not a measured bill, and nothing here says anything about quality or latency.
The calculator runs in your browser. Nothing you type leaves it.
Calculator
Scripts are off in this browser, so the calculator can't run. Every platform rate it uses is in the price table; the paths' figures are in the worked example.
Estimate from list prices, not a client result
How it adds up
| Line | Basis | A month |
|---|
The same minutes on the worked example's paths
From our worked example's arithmetic, with your minutes, phone line and other monthly fees. Each path is an estimate from list prices, not a forecast of what it would cost your team, and a path only counts once it passes quality checks on your calls.
| Path | A month | Per minute | Against your estimate |
|---|
What the calculator does
- Your estimate. Connected minutes times the platform's listed rate, plus what you entered per minute for models on top and the phone line, plus the plan fee and any other monthly fees. On Deepgram's Growth plan, platform usage below $333.33 a month is topped up to that figure: its commitment from $4,000 a year spread over 12 months ESTIMATE, for comparison only, since Deepgram doesn't bill it monthly. Connected minutes include silence, so every rate here applies to all of them. On ElevenLabs only the minutes beyond the plan's included minutes are charged at $0.08.
- Path B. Mistral's voice agent API with Mistral Small 4 at its listed $0.04 a minute, plus your phone line and other monthly fees. Coming from another platform, the model and the voice both change, and you build the phone bridge a platform gives you; that work isn't priced. If you're already on Mistral's API with a larger model, only the model changes, to Small 4.
- Path C. The worked example's self-hosted setup: two AWS c7i.4xlarge and two RunPod L40S always on, at list prices of $0.714 and $1.09 an hour (checked 21 October 2026) for 730 hours ($2,633.84 a month), plus GPT-5.4 by API with its prompt cached ($0.0087 a minute, the worked example's ESTIMATE), plus your phone line and other monthly fees. The worked example sizes those servers for up to 100,000 minutes a month; that capacity is assumed, not tested, so above it the path isn't estimated. Engineering and on-call time aren't included.
- Path A. Caching a repeated prompt depends on how much of your LLM bill is that prompt, so it's only priced on Vapi, the worked example's platform, when you tick the worked example's models, where the LLM line falls from $0.0357 to $0.0087 a minute (both the worked example's ESTIMATE). It's your estimate with that one line changed, so your plan fee, phone line and other monthly fees stay in.
- Not modelled. Per-call rounding (whole minutes, 30- or 60-second pulses), vendors that charge less for silence, contract discounts, volume tiers, promotions, taxes and transferred-call rates. Your other monthly fees are carried into every path as you entered them; paths B and C leave out your platform's plan fee and, on Deepgram's Growth plan, the top-up to its commitment.
The difference is shown in dollars a month from list prices. It isn't a forecast for your team: only your invoices and call logs give your real cost per connected minute, and a path only counts once it passes quality checks on your calls.
Where every number comes from
| Input | Value | Source |
|---|---|---|
| Vapi | $0.05/min hosting; Core plan $29 a month | Price table, from vapi.ai/pricing |
| Retell AI | $0.055/min voice infrastructure | Price table, from retellai.com/pricing |
| ElevenLabs Agents | $0.08/min beyond plan minutes; $990 plan with 12,375 minutes | Price table, from elevenlabs.io/pricing/agents |
| Deepgram Voice Agent API | Pay as you go: $0.075 Standard, $0.065 bring your own text-to-speech, $0.050 bring your own LLM and text-to-speech, $0.163 Advanced. Growth: $0.068, $0.051, $0.041, with a commitment from $4,000 a year LIST, spread here as $333.33 a month ESTIMATE | Price table, from deepgram.com/pricing |
| OpenAI GPT-Live 1 | $0.05/min | Price table, from OpenAI's pricing page |
| Mistral voice agent API | $0.04/min with Mistral Small 4; $0.08/min with a larger model | Price table, from Mistral's Voxtral 3 announcement |
| Bland | $0.14/min Start; $0.12/min Build plus $299 a month | Price table, from bland.ai/pricing |
| Path C servers and cached LLM | $0.714/h c7i.4xlarge and $1.09/h L40S LIST; $0.0087/min cached GPT-5.4 ESTIMATE | Worked example, prices checked 21 October 2026 |
| Worked example stack | Models $0.0634/min: speech-to-text $0.0077 LIST, LLM $0.0357 and text-to-speech $0.0200 ESTIMATE; phone line $0.0096/min ESTIMATE | Worked example |
Our demo prices a different assumed stack, and its same-tier API swap came to $0.1145 a minute there. That figure depends on the demo's stack, so the calculator doesn't use it.
Prices change often. Each vendor links to the page we read; check it before you plan. Spotted a price that has changed? Email poojith@illate.dev and we'll update it.
From list prices to your invoices
The Voice Cost Audit starts from your last 30 days of invoices and your calls, not list prices.