Start here: the real price per minute, checked daily, with receipts
Many pricing roundups copy a vendor's headline price once and never touch it again. VoiceCost rechecks each tracked vendor's official pricing page every day, keeps a dated price history and a changes log, and helps you add up the all-in cost of a minute: platform fee, speech-to-text, LLM, text-to-speech and telephony.
Last updated · prices checked daily
How to use VoiceCost
- Price your call in the calculator. Pick a speech-to-text, LLM, platform, voice and phone line, or switch to "All-in agent platform". Enter your call length and monthly volume to get cost per call and per month.
- Compare two vendors side by side. Pages like Vapi vs Retell or Deepgram vs AssemblyAI line up each product's price. Browse them all on the compare page.
- Check a vendor's page and price history. Every vendor page (for example ElevenLabs) shows each product's price, the unit the vendor uses, a link to the source page, and a price-history line built from every daily check since tracking began on Oct 20, 2026.
- Watch the changes log. When a vendor edits a price, the change lands there with the old and new number and the date we spotted it.
Tip: in the agent table, watch the Scope column. "Platform fee only" means models and phone minutes cost extra. "Bundled" and "All-in" include more, but each vendor's note says what's left out.
What a voice agent call really costs
A voice agent stacks five meters, and each one runs every minute of the call:
- Platform fee: the orchestration layer (for example Vapi, Retell, LiveKit).
- Speech-to-text: turns the caller's voice into text.
- LLM: decides what to say.
- Text-to-speech: speaks the reply (for example ElevenLabs). Billed by characters, so it depends on how much the agent talks.
- Telephony: the phone number and carrier minutes.
Many platforms advertise only the first line. Here's what happens when you add the rest.
Example only, using prices VoiceCost recorded on Oct 28, 2026. Assumptions: an inbound phone call, the agent speaks 50% of the time, and one spoken minute is about 800 characters (Cartesia's pricing page puts it at 750–800). Your stack and call pattern will differ.
| Piece | Choice in this example | $/min | Source |
|---|---|---|---|
| Platform fee | Vapi hosting | $0.0500 | source |
| Speech-to-text | Deepgram Nova-3 streaming | $0.0048 | source |
| LLM | GPT 4.1 mini (per-minute rate from Retell's list) | $0.0128 | source |
| Text-to-speech | ElevenLabs Flash / Turbo ($40/1M chars × 400 chars/min) | $0.0160 | source |
| Telephony | Twilio inbound (rate from Vapi's list) | $0.0080 | source |
| Total per minute | $0.0916 | ||
So the $0.05/min platform fee is about 55% of what a minute actually costs here. A 5-minute call comes to about $0.46, not $0.25. The LLM rate is the per-minute figure Retell publishes for that model; Vapi passes model costs through at cost, so your actual LLM line may differ.
What "checked daily" means
A scheduled job reads 24 vendors' public pricing pages each morning. We currently have verified prices for 20 of them. If a page can't be read or doesn't publish a price, we show Unknown instead of guessing. Every price links to its source and the date it was checked. Full details are in how we collect prices, and the raw numbers are in JSON.
Disclosure: some links (currently ElevenLabs and Vapi) are affiliate links. We may earn a commission at no cost to you. It never changes the prices shown or how they're sorted.