AI Calling Agent Price & Cost in India: Per-Minute Breakdown
The most common question we get after a demo is simple: what is the AI calling agent price in India, per minute? The honest answer is that a quote can mean very different things. One vendor’s ₹3 a minute includes telephony; another’s ₹2 excludes it, and a third charges a platform fee on top of the AI providers you pay separately. Comparing them line by line is the only way to know what you will actually spend.
This post breaks a minute of AI calling into its parts, shows typical ranges for each, explains the common pricing models, and gives you a simple way to estimate your monthly bill. If you are still choosing a vendor, our buyer’s guide to AI calling agents in India covers the non-price criteria.
What you are paying for in one minute of AI calling
Every AI call, whoever runs it, uses the same building blocks. Each has its own cost driver.
| Component | What it does | How it is usually billed |
|---|---|---|
| Speech-to-text (STT) | Turns the caller’s voice into text | Per second or minute of audio processed |
| LLM (the brain) | Decides what to say and which action to take | Per token (roughly, per word read and written) |
| Text-to-speech (TTS) | Turns the reply into a voice | Per character spoken |
| Telephony | The phone line itself: SIP trunk or carrier minutes | Per minute or per pulse, plus number rental |
| Platform or licence | The software that ties it together | Per minute, per month, or per concurrent call |
Two things follow from this. First, a chatty bot costs more than a concise one, because TTS is billed per character and the LLM per token. Second, the cheapest STT is not always cheapest overall: a batch recogniser that adds a pause to every turn can make calls longer.
Typical AI calling agent cost per component
Our gateway documentation keeps a cost and quality guide with per-minute estimates built from providers’ pay-as-you-go list prices. Using its assumptions (a bot that speaks about 470 characters per minute and recognition billed for about 22 seconds per minute of call), the ranges look like this. These are estimates, not quotes, and provider prices change often.
- Speech recognition: from ₹0 for self-hosted VEXYL-STT on your own server, through roughly ₹0.07 to ₹0.16 a minute for batch options such as Groq or Deepgram, to around ₹0.19 for Sarvam streaming and ₹0.59 for Google streaming.
- Speech synthesis: from ₹0 self-hosted, through well under ₹1 a minute for many Indian-language voices, to around ₹5 a minute for premium ElevenLabs voices without caching.
- The LLM: about ₹0.65 a minute in a production setup on Gemini 2.5 Flash, lower on small models, higher on large ones.
Put together, the guide’s “balanced” setup for Indian languages (Sarvam recognition, a Sarvam voice and a LiteBot brain) comes to roughly ₹1.00 to ₹1.78 per minute depending on caching and current Sarvam pricing, before telephony and licence. A fully self-hosted speech setup brings the AI part down to the LLM cost plus your servers.
The caching effect
Greetings, menus, confirmations and fixed phrases repeat on every call. If your system caches synthesised audio, those phrases are paid for once. The docs’ estimates assume two thirds of characters come from cache, which cuts TTS cost by about the same share. Answers built from a caller’s own details are never cached, so the saving depends on your script.
A real number from a live deployment
Estimates are useful, but a real bill is better. In February 2026 a hospital in India running VEXYL handled 156 calls for a total of ₹205.88 in AI costs: ₹0.83 per minute of AI call time. The full breakdown, including which providers were used, is in our post on voice AI cost in India from a live deployment. Treat it as one data point: your call length, language and providers will move the figure.
How vendors package AI calling rates in India
Most AI calling rates in India fall into one of three models.
1. Bundled per-minute rate
You pay one price per minute that covers STT, LLM and TTS, sometimes telephony too. It is simple to budget. The trade-off is that you rarely control which providers are used, and the vendor’s margin sits inside the number. For example, at the time of writing Sarvam publishes voice agent rates starting at ₹3.50 per minute on pay-as-you-go, excluding GST, with telephony billed separately (Sarvam pricing docs). Check the vendor’s current page before you compare.
2. Platform fee plus provider costs
Some platforms show each component separately and add a per-minute platform fee. Bolna’s documentation, for instance, describes call cost as STT, LLM, TTS, telephony and a platform fee, with an option to bring your own provider keys. This model is more transparent but needs a little more arithmetic.
3. Licence plus bring-your-own-keys
You pay for the software (by month or by concurrent calls) and pay the AI providers directly at their list or contract prices. There is no per-minute markup on the AI. This is usually cheapest at volume and gives you the most control, but someone has to manage provider accounts.
Estimating your AI outbound calling cost
For outbound campaigns, cost per connected minute is only part of the story. Unanswered calls, voicemail and retries add telephony charges without much AI usage. Use this simple model:
- Contacts per month multiplied by attempts per contact gives total dial attempts.
- Multiply attempts by your answer rate to get connected calls.
- Multiply connected calls by average minutes per call to get AI minutes.
- AI minutes multiplied by AI cost per minute, plus telephony for all attempts, plus licence, gives your monthly total.
- Divide by successful outcomes (confirmed appointments, qualified leads) to get cost per outcome.
That last figure is the one to compare with your current cost, whether that is telecaller time, missed appointments or unpaid invoices.
A worked example
Say you need 10,000 minutes of AI conversation a month with a balanced Indian-language setup. Using the docs’ estimate of roughly ₹1.00 to ₹1.78 per minute, the AI part comes to about ₹10,000 to ₹17,800 a month. Add your telephony contract and licence. On a bundled ₹3.50-a-minute rate, the same minutes would be ₹35,000 plus GST and telephony. The gap is real, but so is the effort of running your own setup, so weigh both.
Five ways to lower your AI calling cost
- Keep the bot brief. Shorter replies mean fewer TTS characters and fewer LLM tokens, and callers prefer them anyway.
- Turn on TTS caching for greetings and fixed phrases.
- Match the LLM to the task. Reminders and confirmations rarely need the largest model.
- Use batch STT only where callers say little, such as yes/no confirmations. For real conversations, streaming answers faster.
- Self-host speech at high volume, if you have the server capacity.
How VEXYL fits
VEXYL uses the licence plus bring-your-own-keys model. The pricing page lists a free Developer edition (self-hosted, 4 concurrent calls), Professional at $49 a month (25 concurrent calls) and custom Enterprise plans. You connect your own Sarvam, Google, Deepgram, Groq, Gemini, ElevenLabs, Azure or other provider accounts, or use the open-source VEXYL-STT and VEXYL-TTS to avoid per-minute speech fees.
The gateway records real per-call costs in its billing records and call.ended webhook, so you see actual spend per call rather than an estimate. That is how the ₹0.83 figure above was measured.
FAQ
What is the average AI calling cost in India per minute?
Published rates from Indian vendors commonly sit in the low single-digit rupees per minute, but what is included differs. With bring-your-own-keys setups, the AI part can be around ₹1 a minute or less, before telephony and licence.
Is telephony included in AI calling agent prices?
Often not. Many vendors bill telephony separately per minute or pulse, plus number rental. Always ask.
Why do AI outbound calling costs vary so much between campaigns?
Answer rates, retries, call length and how much the bot talks all change the bill. A campaign with many unanswered attempts spends more on telephony per successful outcome.
Is self-hosting always cheaper?
Not always. It removes per-minute speech fees and vendor markups, but you pay for servers and someone’s time. It tends to pay off at higher, steady volumes or when data rules require it anyway.
Does GST apply to AI calling services?
Vendors in India typically quote prices excluding GST. Confirm with your vendor and accountant how it applies to your invoices.
Want a cost estimate for your own call volumes? Contact the VEXYL team with your expected minutes and languages, or try the free Developer edition and read the real per-call costs yourself.