Costs per minute

Your assistant can tell you what a minute of calling will cost before it creates an agent or starts a campaign — worked out the way the agent builder's cost bar works out its estimate.

Calls an assistant makes are billed exactly like any other call in Telenow: by what the call actually used, from your organization's wallet, under your spend and usage limits (Usage & billing). The estimate is for planning.

Ask for it

You: What would a minute of the clinic agent cost? And with the cheaper voice?

The assistant uses estimate_cost, either for an agent you have:

{ "name": "estimate_cost", "arguments": { "agent_id": "5d2e…" } }

…or for a setup you are still planning — anything left out is your organization's default for the country:

{
  "name": "estimate_cost",
  "arguments": {
    "voice": { "provider": "cartesia" },
    "system_prompt": "You are calling clinics to ask about walk-in hours…",
    "country": "IN"
  }
}

{
  "currency": "USD",
  "per_minute": {
    "model": 0.0002,
    "speech_to_text": 0.0077,
    "voice": 0.024,
    "telephony": 0.006,
    "platform_fee": 0.0115,
    "total": 0.0494
  },
  "stack": {
    "model": { "provider": "openai", "model": "gpt-4o-mini" },
    "speech_to_text": { "provider": "deepgram", "model": "nova-3" },
    "voice": { "provider": "cartesia", "model": null },
    "carrier": "plivo",
    "own_carrier": false
  },
  "not_priced": [],
  "note": "An estimate from average call usage (the agent builder's assumptions); a chatty call costs more, a short exchange less. …"
}

Assistant: About $0.05 a minute with that voice — half of it the voice. A 2-minute call to 100 people would be about $10.

What goes into it

Part	How it is priced
Model	The model's catalogue price for the tokens of an average minute of conversation — your prompt (its length matters) re-sent each turn, plus what the agent says. A model with cached prompts is priced at its cache rate.
Speech-to-text	The provider's (or model's) per-minute rate for the whole minute.
Voice	The voice's price per character for what an agent says in an average minute.
Carrier	The per-minute rate of the carrier behind the number the calls come from — `from_number`, or else the first of the assistant's numbers. If that carrier has no catalogue price, the costliest carrier's rate is used — an estimate should not under-promise.
Platform fee	Your organization's platform fee — the platform's default or your account's own rate, plus any pricing rules that apply to this setup.

The averages — how many words a minute, how many turns — and the formula are the agent builder's own. The builder's cost bar can still show a different figure: it also prices what it measures in the builder — the platform's own instructions sent with every turn, a flow's individual steps, a model's hosting route — which the assistant's estimate does not see.

What counts as zero

Your own vendor keys. A part that runs on your own API key (bring-your-own-key) is billed to you by that vendor, not by Telenow, so it is 0 here.
Your own carrier. A number on your own carrier (Carriers (BYOC), SIP trunking) puts the call minutes on that carrier's bill, so the carrier part is 0.

What it does not price

Realtime voice models. An agent that talks through a speech-to-speech model is priced differently; the assistant says so and points you to the agent builder, which shows its cost.
A model, speech-to-text or voice the catalogue has no price for is listed in not_priced and counted as 0 — the assistant tells you the estimate leaves it out.

Real calls cost what they use

A chatty call costs more than the average minute suggests, a short exchange less. Your call history shows what each call actually cost (Calls, recordings & insights).