voicecosts

THE VOICE AGENT BUDGET PLANNER

A minute has a price.
An outcome has a cost.

See the moving parts behind your AI voice bill. Model a call, pressure-test success rates, and take a clearer budget into your next build.

No accountInputs stay in your browserUSD estimates
01

Build your scenario

Start with a billing route, not a provider ranking.

Published voice session rate; backend reasoning and tools are separate. No separate STT/TTS assumed.

A success is the task you define: for example, a resolved support request. This is not a provider benchmark.

02

Account for the components

Blank means unknown. Enter 0 only when a cost does not apply.

Published: $0.05 / active minute, billed per second. Verified October 9, 2026.

Your effective rate from usage or a pilot.

0 in this route: native audio is included.

Your country, direction and call-leg rate.

Numbers, hosting, support and other fees.

Assumes one carrier leg per session and identical durations. Model multiple legs separately. Voice/model minutes remain unrounded.

A BETTER BUDGET CONVERSATION

Three things a headline rate cannot tell you.

01 /

What counts as a minute?

Use the full active session, including silence and backend waiting. GPT-Live bills per second; a carrier may round each leg. An average-length call is only a planning approximation.

02 /

What is still missing?

A hosting fee is not a complete voice stack. Add architecture-specific models, tools, speech, carrier, phone numbers and support. Blank fields remain listed as excluded costs.

03 /

Did the task get done?

Monthly subtotal divided by successful sessions gives the modeled cost per outcome. A low minute price can still be expensive when calls fail. Validate your success rate with a real pilot.

See the formulas and planning limits

Active minutes = sessions × active seconds ÷ 60. Known subtotal = active minutes × (voice/hosting + model/tools + separate speech rates) + carrier billed minutes × carrier rate + fixed monthly costs. Blank components are excluded, not assumed free.

When carrier rounding is on, billed minutes = sessions × ceil(seconds ÷ 60). Real invoices round each actual call leg individually; this simplified planner assumes equal session lengths and one leg. Cost per success = known subtotal ÷ (sessions × success percentage ÷ 100). Zero sessions or zero success produces “Not defined,” not a misleading zero.

Effective per-minute model rates must come from your own pilot or invoice; this tool does not convert token prices into minutes. GPT-Live's 15-second WebRTC initialization charge is credited against running duration, so do not add 15 seconds to each completed call. Failed initialization, retries, minimums and multi-leg transfers need separate allowance.

CHECK THE BILLING, THEN THE BUDGET

Sources you can inspect.

Rates checked October 9, 2026. Published defaults cover only the stated component. All other numbers are your assumptions. Recheck provider terms before spending.

OpenAI · GPT-Live-1 pricing ↗

$0.05 per active session minute, per-second billing. Backend model/tool charges are separate.

Vapi · usage-only hosting ↗

$0.05 per hosting minute. Models pass through at cost; speech and carrier depend on your stack. This preset is not a GPT-Live-through-Vapi quote.

OpenAI · voice cost mechanics ↗

Session duration, initialization and separate backend costs.

Twilio · example carrier pricing ↗

Direction, country and call legs matter. No carrier rate is preselected.

Your inputs are your business.

No login, lead form or saved scenarios. Calculations and pinned baselines stay on this page and disappear on reload. Copying uses your device clipboard.

Optional Google Analytics measures page views and fixed actions only: calculate, switch template, pin, copy and open a pricing source. We do not send your input values, scenario text, URL query, or referrer. Google receives ordinary network information when enabled. Consent lasts for this page visit.

Optional analytics is off.

The calculator works either way.