Per-Minute Billing for Voice AI: What Actually Gets Charged
Founder of Bolti, writing about voice AI for Indian businesses.
Per minute billing voice ai is a pricing model where you pay only for the exact duration of connected phone calls handled by an artificial intelligence agent. Bolti, a voice AI platform for building production-ready multilingual phone agents, offers a flat ₹6/minute pay-as-you-go rate with 50 free minutes to start, making it easy to predict and control your telephony costs.
What is per-minute billing for voice AI?
Per minute billing voice ai charges you only for the active duration of connected calls, calculated from the exact second the recipient answers until either party hangs up. You do not pay for setup, ringing time, busy signals, or failed dial attempts.
The key word is connected. Unlike legacy platforms or complex setups that charge for dialing attempts, Bolti starts the billing clock only when the called party picks up. If an outbound call rings for 25 seconds and goes unanswered, it costs you nothing.
What the per-minute rate actually covers
A single flat rate of ₹6/minute bundles several underlying technical layers that run continuously in a streaming loop:
- Telephony Leg: Carrying the call over PSTN or SIP trunks (using carriers like Twilio, Exotel, or Plivo).
- Speech-to-Text (STT): Transcribing the caller's voice in real time using low-latency engines like Deepgram, AssemblyAI, or Fennec (optimized for Indian accents).
- LLM Inference: The cognitive model (like GPT-4o, Gemini, or Groq-hosted models) reading the transcript and deciding the next response.
- Text-to-Speech (TTS): Synthesizing the agent's text response back into natural audio via Cartesia or ElevenLabs.
How much does voice AI cost per minute in 2026?
In 2026, voice AI costs range from ₹5 to ₹25 per minute depending on your choice of LLM, TTS quality, and platform markups. Bolti simplifies this with a flat ₹6/minute rate that bundles all STT, LLM, TTS, and telephony costs with no monthly subscription fees.
When evaluating competitors like Bolna AI or Ringg AI, you will generally encounter four distinct billing structures:
| Billing Model | How It Works | Risks & Hidden Costs |
|---|---|---|
| Flat Per-Minute (Bolti) | One predictable rate (₹6/min) covering STT, LLM, TTS, and telephony. | None; ideal for budgeting and scaling. |
| Component-Based | Billed separately for STT audio, LLM input/output tokens, and TTS characters. | Costs spike unpredictably during long conversations or complex prompts. |
| Subscription + Overage | Fixed monthly seat fee plus a lower per-minute rate for calls. | Underutilized seats waste your budget; high upfront commitment. |
| Per-Call Flat Rate | A single fixed fee per connected call, regardless of duration. | Extremely expensive for short calls (e.g., 30-second payment reminders). |
What exactly counts as a billable second?
Every second from the moment the call connects to the moment it disconnects is billable, including pauses, silence, and API latencies. Non-connected ring time, failed calls, and post-call processing are completely free.
To avoid unexpected charges, keep these factors in mind:
- Continuous Voice Activity Detection (VAD): The voice pipeline does not pause when neither party is speaking. Bolti's VAD and turn-detection algorithms run continuously to handle interruptions and stream audio, meaning silent pauses are still billable.
- Tool Call Latency: When your agent queries an external database, CRM, or booking API, the call remains active. If your CRM takes 3 seconds to fetch data, those 3 seconds are billable. Keep your APIs fast.
- What is NOT billable: Ringing time, busy tones, unanswered calls, and post-call webhook execution or recording processing.
How do provider choices in the settings tabs affect your bill?
While Bolti maintains a flat ₹6/minute rate regardless of your configuration, your choice of STT, LLM, and TTS providers directly impacts call duration, which determines your total bill. Faster, more concise providers keep calls short and costs low.
When configuring your agent in the Bolti dashboard, pay close attention to these tabs:
- The Voice Tab: Here you can select from a grid of voice cards (like Aria, Marcus, or Anushka) powered by Cartesia or ElevenLabs. Choosing a low-latency model like Cartesia's Sonic-3 reduces turn-taking delays, keeping the conversation moving quickly.
- The Speech Tab: Choosing an optimized provider like Fennec for Indian languages (Hindi, Marathi, Tamil, Telugu, Gujarati) ensures faster, highly accurate transcriptions, preventing callers from repeating themselves and lengthening the call.
- LLM & Prompt Tuning: Concise system prompts prevent the agent from generating long-winded responses. Every extra sentence the agent speaks increases the TTS generation time and overall call duration.
See Bolti use cases to understand how different configurations perform across outbound sales, HR screening, and support.
How are batch campaigns and bulk calling billed?
Batch campaigns—including recurring cron reminders and bulk list dialers—are billed at the same flat ₹6/minute rate for connected calls. You are never charged for the scheduling infrastructure, pacing, or failed call attempts within a campaign.
Whether you are running bulk campaigns for Q3 follow-ups or setting up recurring reminders, the billing remains completely transparent:
- Bulk Campaigns: Upload a list of contacts, and Bolti's dispatchers will dial them at a controlled rate. You only pay for the calls that connect.
- Recurring Campaigns: Scheduled calls (like a daily standup reminder at 9:00 AM IST or weekly payment nudges) use the same per-second connected billing.
- No Platform Overhead: The singleton scheduler, materializer pipeline, and webhook dispatchers run completely free. You only pay for the actual talk time on the PSTN/SIP carrier leg.
How to estimate your monthly voice AI budget
To estimate your monthly budget, multiply your expected daily call volume by the average call duration in minutes, the ₹6 flat rate, and 30 days. This gives you a clear baseline with no hidden platform fees.
Monthly Cost = (Daily Calls × Avg Duration in Minutes × ₹6) × 30
Example calculation for an Indian SMB: A Bengaluru-based logistics company runs 300 automated delivery confirmation calls per day. The average call duration is 1.5 minutes.
- Daily Minutes: 300 calls × 1.5 minutes = 450 minutes/day
- Daily Cost: 450 minutes × ₹6 = ₹2,700/day
- Monthly Cost: ₹2,700 × 30 days = ₹81,000/month
Compare this to human agents or complex component-based pricing from competitors like Bolna AI, where unpredictable token usage makes budgeting nearly impossible. For high-volume operations running over 10,000 calls per month, you can request custom volume pricing by reaching out to the team at the Bolti contact page.
Try Bolti for your business calls
Spin up your first multilingual voice agent in under 10 minutes and see exactly how per-second billing works in practice. Bolti offers a free trial with 50 minutes of call time so you can test our sub-second latency, Indian language models, and CRM integrations with zero upfront commitment. Once you are ready to scale, our transparent ₹6/minute pay-as-you-go pricing ensures you only pay for what you use.
Frequently Asked Questions
Does a 30-second call cost less than a 1-minute call?
Yes. Bolti bills on a per-second prorated basis. At our flat ₹6/minute rate, a 30-second connected call costs exactly ₹3.00. You are never rounded up to the next full minute.
What if the caller hangs up immediately or it goes to voicemail?
If the call connects (e.g., a person or a voicemail system answers), you are billed for the exact duration it was live. If the call does not connect (busy signal, network error, or unanswered ring), you pay nothing.
Are inbound and outbound calls billed the same way?
Yes. Whether a customer dials your inbound Bolti number or your agent triggers an outbound campaign, billing is calculated identically at ₹6/minute based on connected call duration.
Do I get charged for the time the AI takes to think?
Yes, the call remains connected while the LLM processes and the TTS synthesizes the response. However, Bolti uses ultra-low latency models (like Groq and Cartesia) to keep turn-taking under 800ms, minimizing idle billable time.