Voice calls from $0.06/min. Chat and messaging billed by the token. You control the cost.
No seat licenses. No bundled black box. Calls are priced per minute by the models you pick; chat, WhatsApp, SMS, RCS, email and API conversations are priced by the AI tokens they actually use — and you see both before you launch.
The entry-level stack
Pick the cheapest core models and voice calls start at about $0.06/min. This is the live per-minute stack you configure in the agent builder, shown component by component before you launch.
- STT — Soniox Realtime$0.005/min
- LLM — Gemini 2.5 Flash$0.020/min
- TTS — Smallest Lightning v3.1$0.035/min
Agent builder
One agent · every channel · your models
WhatsAppVoice pipeline
STT
SonioxRealtime
LLM
Google Gemini2.5 Flash
TTS
Smallest AILightning v3.1
Chat and messaging are billed by tokens, once per conversation.
Optional add-ons
Only pay for these when you switch them on. Nothing here is bundled into the base rate.
Advanced noise cancellation
$0.004/minCleaner audio on noisy lines and outdoor calls.
Post-call transcription
$0.007/minFull searchable transcript generated after each call.
LLM prompt overage
$0.004/minOnly applies when your prompt runs over 5,000 tokens. The first 5,000 prompt tokens are included in every call.
- STT — Soniox Realtime$0.005/min
- LLM — Gemini 2.5 Flash$0.020/min
- TTS — ElevenLabs Flash v2.5$0.10/min
Premium voice example. Swaps TTS to ElevenLabs Flash v2.5 ($0.10/min) for a more expressive voice. Total lands around $0.14–0.15/min depending on models.
Premium voice example
When the voice is the brand, upgrade the TTS. Swapping in ElevenLabs Flash v2.5 ($0.10/min) — while keeping the same STT and LLM — brings a typical call to around $0.14–0.15/min.
You decide per agent: cost-efficient models for high-volume operational calls, premium voices for customer-facing brand moments.
Token-based pricing for every text channel
Website and app chat, WhatsApp, SMS, RCS, email and the API channel are billed on the AI tokens each conversation actually uses — input, cached input and output — at the rate of the language model you choose. You only pay for the thinking the agent does.
Pay for tokens, not messages
Each AI reply is priced on its real input, cached input and output tokens for the model you pick. Short replies on SMS cost less than long emails — automatically.
One line per conversation
A conversation's usage is collected and billed as a single line when it finishes (or daily for long threads) — not one charge per message. Every AI attempt is billed once, never twice.
Cached prompts cost less
Your agent's instructions and knowledge are cached between replies, so cached input is billed at the model's lower cached rate.
See the cost before launch
The agent builder shows the expected cost per AI reply and per conversation for each channel, then switches to your agent's measured usage once it has real conversations.
Your team's replies are free
Messages your teammates send from the inbox carry no platform fee — hand-offs never become a cost centre.
Caps you control
Set a daily spend cap with an alert before you reach it, prepaid wallets, and per-contact reply limits. Test conversations never count toward live budgets.
Messaging provider fees — WhatsApp conversation charges, SMS segments, RCS and email sending — are billed by your own provider account at their rates, not by Callaro.
Speech-to-speech models for the most natural calls
Instead of the STT → LLM → TTS chain, run an agent on a realtime speech-to-speech model for lower perceived latency and more natural turn-taking. Priced per minute and shown in the builder, like every other model.
OpenAI GPT-Live-1
Full-duplex speech-to-speech from OpenAI (also available via Azure OpenAI), delegating reasoning and tools to a text model so the agent keeps acting mid-call.
Google Gemini Live
Gemini 2.5 Flash native-audio realtime model with natural voices, barge-in and multilingual speech.
You choose the models
Run a speech → language → voice chain with each layer swappable per agent, or a single speech-to-speech model. Mix and match to hit the accuracy, latency, language coverage and cost that fits the use case.
Speech-to-text (STT)
How the agent hears and transcribes the caller.
Text-to-speech (TTS)
The voice your customers hear — from cost-efficient to studio-grade.
Language model (LLM)
The reasoning that runs the conversation and decides the next action.
Speech-to-speech (realtime)
One realtime model that listens and speaks directly — lowest perceived latency, most natural turn-taking.
Estimate your savings
Drag the sliders to see monthly Callaro cost at $0.06/min against what the same calls cost with people.
Assumes a fully-loaded US agent cost at 65% productive occupancy (the rest is after-call work, breaks and idle time), over 21 working days/month at $0.06/min. Premium voices and add-ons cost more.
Human cost / month
$10,769
Callaro / month
$1,008
Estimated monthly savings
$9,761
About 91% lower than doing these calls with people.
4,200 calls/month · 16,800 minutes
What's included
Everything below is part of the platform — no per-seat fees, no add-on charges for the dashboard.
Pricing questions
How is the per-minute price calculated?
Every call is billed by the models you choose: one STT rate, one LLM rate and one TTS rate, added together per minute. The entry-level stack (Soniox + Gemini 2.5 Flash + Smallest Lightning v3.1) works out to about $0.06/min. Swap in premium models and the per-minute rate goes up accordingly — you always see the total before you launch.
What counts as an add-on?
Advanced noise cancellation ($0.004/min) and post-call transcription ($0.007/min) are optional and only cost when you enable them. LLM prompt overage ($0.004/min) applies only when a call's prompt exceeds 5,000 tokens — the first 5,000 are included.
Why would I pick a premium voice?
For customer-facing brand calls where the voice matters, you can upgrade TTS — for example to ElevenLabs Flash v2.5 at $0.10/min — bringing a typical call to around $0.14–0.15/min. For high-volume operational calls, the entry-level stack keeps costs low.
How is chat and messaging priced?
Website and app chat, WhatsApp, SMS, RCS, email and API conversations are billed on the AI tokens each conversation uses — input, cached input and output — at the rate of the language model you choose, and billed as one line per conversation. The agent builder shows the expected cost per AI reply and per conversation for each channel before you launch. Your messaging provider bills its own fees (for example WhatsApp conversation charges or SMS segments) on your account.
Can I use speech-to-speech models like GPT-Live or Gemini Live?
Yes. Alongside the STT → LLM → TTS chain, you can run an agent on OpenAI GPT-Live-1 (directly or via Azure OpenAI) or Google Gemini Live for lower perceived latency and more natural turn-taking. These are priced per minute and shown in the agent builder before you launch.
Do my team's replies in the inbox cost anything?
No. Messages your teammates send when they take over a conversation carry no platform fee.
Are there seat or user fees?
No. There are no per-seat or per-agent licenses. You pay for the minutes and tokens you use, and every channel, the inbox, dashboards, transcripts, follow-ups and integrations are part of the platform.
Do you offer volume or enterprise pricing?
Yes. As your volume grows we work with you on tailored pricing and dedicated support. Talk to sales for a quote built around your projected minutes.
Running high volume or need enterprise controls?
We'll build a quote around your projected minutes.
Talk to salesSee your cost per outcome, live.
Book a demo and we'll configure a stack for your use case and show the exact per-minute price.
Book a demo