---
title: "AI Voice Agent Pricing India 2026: Cost Per Minute | Vobiz"
description: "AI voice agent pricing in India for 2026: a 3-minute call priced in INR across telephony, STT, LLM, TTS and platform fees, plus hidden costs to watch."
canonical: https://www.vobiz.ai/resources/blogs/buyers-guide/ai-voice-agent-cost-per-minute-india/
---
[Blogs](https://www.vobiz.ai/resources/blogs/)Buyer's Guide

# AI Voice Agent Pricing in India 2026: Real Cost Per Minute in INR

What an AI voice agent costs per minute in India, broken into telephony, STT, LLM, TTS and platform fees, with a worked 3-minute call in rupees.

![AI Voice Agent Pricing in India 2026: Real Cost Per Minute in INR](https://www.vobiz.ai/blog-thumbs/ai-voice-agent-cost-per-minute-india.jpg)

![Team Vobiz](https://www.vobiz.ai/favicon-192.png)Team Vobiz

10 min read19 November 2026

Share this article

On this page

Subscribe

For newsletter and updates

An AI voice agent in India costs roughly **₹2 to ₹11 per minute** in November 2026, once you add up telephony, speech-to-text (STT), the language model (LLM), text-to-speech (TTS) and the orchestration platform fee. A lean self-hosted stack with Indian-language models sits at the bottom of that range; a hosted platform with premium voices sits at the top. The per-minute sticker is only part of it: unanswered calls, number rental, long-call token growth and the dollar-rupee rate decide what you actually pay per conversation.

Below: the five cost layers, a 3-minute call priced in rupees from each vendor's own pricing page, and the costs a per-minute quote leaves out.

## What goes into the cost of an AI voice agent call?

Most production voice agents in 2026 are still "cascaded": the caller's audio is transcribed, the text goes to an LLM, and the reply is turned back into speech. Each step is billed separately, by a different meter.

| Layer | What it does | How it's usually billed |
| --- | --- | --- |
| Telephony | Carries the call over the phone network (PSTN or SIP) | Per minute, plus monthly number rental |
| Speech-to-text | Turns the caller's speech into text in real time | Per minute of audio streamed |
| LLM | Decides what to say and which tools to call | Per million input and output tokens |
| Text-to-speech | Speaks the reply | Per 1,000 characters |
| Orchestration platform | Turn-taking, interruptions, tool calls, call flow, logs | Per minute (hosted) or your own compute (self-hosted) |

Speech-to-speech models such as OpenAI's realtime models fold STT, LLM and TTS into one meter. OpenAI lists gpt-realtime-mini at $10 per million audio input tokens and $20 per million audio output tokens, as listed on its pricing page in November 2026. That is billed per audio token rather than per minute, so you need to measure your own calls before comparing it with a cascaded stack.

## Bundled vs unbundled voice AI pricing

Vendors package these layers in two ways. Plivo's June 2026 post, [Voice AI Agent Pricing Models in 2026: Bundled vs Unbundled Costs](https://www.plivo.com/blog/voice-ai-agent-pricing-models-bundled-vs-unbundled-costs/), defines them well: bundled pricing "combines several cost layers into one rate, plan, or quote", while unbundled pricing "separates platform usage from the underlying model, voice, and channel costs."

-   **Bundled** is easier to forecast. Deepgram's Voice Agent API is listed at $0.075/min (Standard) and $0.163/min (Advanced), and ElevenLabs lists its Agents platform at $0.08/min, with $0.16/min burst pricing above your concurrency limit, as listed on their pricing pages in November 2026. Always check which layers a bundle includes and which (often the LLM or telephony) are passed through separately.
-   **Unbundled** gives you line-item control. Vapi lists a $0.05/min hosting fee with models passed through at cost, and Retell lists "Retell Voice Infra" at $0.055/min with LLM, TTS and telephony priced separately, as listed on their pricing pages in November 2026. Retell's own page quotes an overall pay-as-you-go range of $0.07 to $0.31 per minute.

Bundled suits pilots that want one number. Unbundled pays off at volume, when you want to swap the expensive layer without changing everything else.

## Worked example: cost of a 3-minute AI call in INR

Here is one typical call, priced three ways. All dollar prices are converted at **₹94 per US dollar**, close to the rate the rupee traded at in mid-November 2026 ([94.13 on 16 November](https://hdfcsky.com/news/rupee-rises-6-paise-to-94-13-against-us-dollar-in-early-trade-november-16-2026)).

### Assumptions for the call

-   **Duration:** 3 minutes, connected and answered.
-   **Turns:** 12 agent turns.
-   **LLM tokens:** a 1,500-token system prompt, with each turn adding about 80 tokens of conversation history. Because the full context is re-sent every turn, input tokens are 12 × 1,500 + 80 × (0 + 1 + … + 11) = 18,000 + 5,280 = **23,280 input tokens**. Output is 12 × 50 = **600 tokens**. No prompt caching.
-   **TTS:** the agent speaks for about half the call, roughly **1,500 characters**.
-   **STT:** billed on the full 3 minutes of streamed audio.

### Component prices used (as listed in November 2026)

| Component | Price as listed in November 2026 | Source |
| --- | --- | --- |
| Vobiz India telephony | ₹0.42/min, inbound and outbound | [Vobiz pricing](https://www.vobiz.ai/pricing/) |
| Sarvam speech-to-text (real-time) | ₹0.49/min, pay as you go | [Sarvam API pricing](https://www.sarvam.ai/api-pricing) |
| Deepgram Nova-3 Multilingual (streaming) | $0.0058/min (labelled promo pricing) | [Deepgram pricing](https://deepgram.com/pricing) |
| OpenAI gpt-4o-mini | $0.15 / $0.60 per 1M input / output tokens | [OpenAI pricing](https://developers.openai.com/api/docs/pricing) |
| OpenAI gpt-4.1-mini | $0.40 / $1.60 per 1M input / output tokens | [OpenAI pricing](https://developers.openai.com/api/docs/pricing) |
| OpenAI gpt-5.4-mini | $0.75 / $4.50 per 1M input / output tokens | [OpenAI pricing](https://developers.openai.com/api/docs/pricing) |
| Sarvam Bulbul V4 Flash TTS | ₹2.00 per 1,000 characters | [Sarvam API pricing](https://www.sarvam.ai/api-pricing) |
| Deepgram Aura-2 TTS | $0.030 per 1,000 characters | [Deepgram pricing](https://deepgram.com/pricing) |
| ElevenLabs Multilingual v2 TTS | $0.08 per 1,000 characters | [ElevenLabs API pricing](https://elevenlabs.io/pricing/api) |
| Vapi hosting fee | $0.05/min | [Vapi pricing](https://vapi.ai/pricing) |

### The arithmetic

| Layer | Lean (self-hosted, Indian models) | Mid (hosted platform) | Premium (hosted, premium voice) |
| --- | --- | --- | --- |
| Telephony (Vobiz) | 3 × ₹0.42 = **₹1.26** | **₹1.26** | **₹1.26** |
| STT | Sarvam: 3 × ₹0.49 = **₹1.47** | Deepgram: 3 × $0.0058 × 94 = **₹1.64** | Deepgram: **₹1.64** |
| LLM | gpt-4o-mini: (23,280 × $0.15 + 600 × $0.60) ÷ 1M × 94 = **₹0.36** | gpt-4.1-mini: (23,280 × $0.40 + 600 × $1.60) ÷ 1M × 94 = **₹0.97** | gpt-5.4-mini: (23,280 × $0.75 + 600 × $4.50) ÷ 1M × 94 = **₹1.90** |
| TTS | Bulbul V4 Flash: 1.5 × ₹2.00 = **₹3.00** | Aura-2: 1.5 × $0.030 × 94 = **₹4.23** | ElevenLabs v2: 1.5 × $0.08 × 94 = **₹11.28** |
| Platform fee | Open-source orchestration (e.g. LiveKit Agents, Pipecat): **₹0** plus your servers | Vapi: 3 × $0.05 × 94 = **₹14.10** | Vapi: **₹14.10** |
| **Total per 3-minute call** | **₹6.09** + compute | **₹22.20** | **₹30.18** |
| **Per minute** | **≈ ₹2.03** + compute | **≈ ₹7.40** | **≈ ₹10.06** |

So the honest answer to "what does voice AI cost per minute in India" is **about ₹2 to ₹11**, with most hosted-platform builds landing between ₹7 and ₹11. Treat these as ranges, not quotes. Your prompt length, how much the agent talks, the voice you pick and any volume discount will move each line.

Two things stand out. First, in the mid scenario the platform fee is ₹14.10 of ₹22.20, about 64% of the call. Second, telephony is the smallest or second-smallest line in every column. Picking a cheaper carrier helps, but the larger savings usually come from the platform fee and the TTS voice.

## Hidden costs that per-minute quotes leave out

### Unanswered calls, ring time and voicemail

Outbound campaigns pay per conversation, not per minute. Carriers generally bill answered time, but check two things with every vendor: whether the orchestration platform starts its meter at dial or at answer, and the billing increment (per second or rounded up to a full minute).

Voicemail is the bigger leak. If an agent talks to a voicemail greeting for 30 seconds on 200 out of 1,000 dials, that's 100 billed minutes, about ₹740 at the mid-stack rate, spent on no one. Answering machine detection ends or redirects those calls early.

### Retries

A typical outbound flow retries no-answers two or three times. Each retry that connects to voicemail or a busy tone costs the minimum billed unit on every metered layer. Budget for attempts, not just contacts.

### Number rental

Numbers are a fixed monthly cost spread across your minutes. On Vobiz, 140-series, 1600-series and 92-series mobile numbers are ₹1,000/month each. Twenty numbers is ₹20,000/month; over 50,000 minutes that adds ₹0.40/min. Over 5,000 minutes it adds ₹4/min, which can double a lean stack's cost.

### Spam-flagged numbers and answer rates

If your numbers get flagged as spam by caller-ID apps or handset filters, fewer people pick up. Your per-minute cost stays the same, but the cost per completed conversation goes up because you burn more attempts, retries and numbers to reach the same people. Rotating numbers sensibly, using the right number series for the use case and watching answer rates per number all protect this. Our [number health ebook](https://www.vobiz.ai/resources/ebooks/number-health-ebook/) goes into this in detail.

### LLM token growth on long calls

Because most agents re-send the full conversation every turn, LLM cost grows faster than call length. Using the same assumptions as above, a 6-minute call with 24 turns sends 24 × 1,500 + 80 × (0 + 1 + … + 23) = **58,080 input tokens**, 2.5 times the 3-minute call for twice the duration. On gpt-4.1-mini that's about ₹2.36 per call instead of ₹0.97. Tool-call results (CRM lookups, order details) pasted into context make this worse.

### USD to INR movement

Most AI vendors bill in dollars. In the mid scenario, everything except telephony (about ₹20.94 of ₹22.20) is dollar-denominated, so a ₹2 move in the exchange rate changes your bill by roughly 2%. Your bank's foreign-exchange markup on international card payments sits on top of that. Costs billed in rupees, such as Vobiz telephony and Sarvam's models, don't move with the exchange rate.

## How to cut AI voice agent cost per minute

1.  **Attack the platform fee first.** At volume, moving from a hosted per-minute platform to open-source orchestration (LiveKit Agents, Pipecat) removes the largest line, in exchange for running and monitoring your own servers. If you stay hosted, negotiate once you have steady volume.
2.  **Pick the TTS voice deliberately.** In our example, premium TTS cost more than three times a fast voice model. Test whether callers in your use case actually notice the difference, especially for Hindi and other Indian languages, where Indian-built models such as Sarvam's Bulbul are an option.
3.  **Use prompt caching and trim context.** Cached input tokens are listed far cheaper than fresh ones on OpenAI's pricing page. Keep the system prompt stable so it caches, summarise old turns, and don't paste whole API responses into context.
4.  **Right-size the LLM.** Small models handle most scripted flows (reminders, confirmations, qualification). Route only the hard turns to a larger model.
5.  **Detect voicemail and cap duration.** Use answering machine detection, set a maximum call length, and end calls cleanly once the goal is met.
6.  **Protect answer rates.** Monitor answer rate per number, rotate before numbers burn, and use the right series for the job.
7.  **Bring your own telephony.** Vapi lists SIP and its own telephony transport as free, and Retell lists no charge for SIP trunking or custom telephony, as listed on their pricing pages in November 2026. That lets you choose an Indian carrier billed in rupees instead of a marked-up default.
8.  **Model the whole month, not one minute.** Multiply cost per minute by expected minutes, then add numbers, retries and failed attempts. Compare stacks on cost per completed conversation.

## How to run the telephony layer on Vobiz

Vobiz handles the telephony line in the tables above: Indian and international numbers, SIP trunks connected to Airtel, Jio and VIL, and real-time audio streaming to your agent.

-   **Pricing:** a flat ₹0.42/min inbound and outbound for India, billed in INR with a GST invoice, pay-as-you-go with no setup fee or minimum commitment. Enterprise pricing is available above about 50,000 minutes a month. See [pricing](https://www.vobiz.ai/pricing/).
-   **Works with your agent platform:** native integrations for Vapi, Retell AI, ElevenLabs, LiveKit, Pipecat, Bolna, Ultravox and OpenAI Realtime over [SIP trunking](https://www.vobiz.ai/products/sip-trunking/) or WebSocket audio streaming through the [Voice API](https://www.vobiz.ai/products/voice-api/). Our guide to the [telephony layer for AI voice agents](https://www.vobiz.ai/resources/blogs/voiceai-infrastructure/telephony-layer-for-ai-voice-agents/) walks through each setup.
-   **Cost controls:** answering machine detection, managed number reputation and spam-risk detection, and 43-field CDRs so you can reconcile billed minutes against your platform's logs.
-   **Numbers:** 140-series, 1600-series, 92-series mobile, toll-free and local numbers, with hosted KYC that completes in minutes.

[Sign up on the console](https://console.vobiz.ai/) and you can have a first call live in about 10 minutes. The [docs](https://www.vobiz.ai/docs/introduction) cover the rest.

Where another option fits better: if you're early and want a single bill with no integration work, a fully bundled agent product is simpler, even if it costs more per minute. Vobiz is the better fit once you want to own your stack, pay for India calling in rupees and keep telephony as a separate, cheap line item.

## FAQ

### What is the cost per minute of an AI voice agent in India?

About ₹2 to ₹11 per minute in November 2026, adding up telephony, STT, LLM, TTS and the platform fee. Most hosted-platform builds land between ₹7 and ₹11. Self-hosted stacks with Indian models can come in near ₹2 to ₹3 per minute plus server costs.

### Which part of the voice AI stack costs the most?

On hosted platforms, usually the per-minute platform fee, followed by TTS. In our mid-range example the Vapi fee was about 64% of the call. Telephony and the LLM were the smallest lines for a 3-minute call.

### Is bundled or unbundled voice AI pricing cheaper?

It depends on volume. Bundled pricing is simpler to forecast and good for pilots. Unbundled pricing lets you replace the most expensive layer, which tends to be cheaper at scale, but you manage more vendors.

### Why does my AI calling cost in India look higher than the per-minute estimate?

Usually because of attempts that don't become conversations: voicemail, retries, low answer rates on spam-flagged numbers, plus monthly number rental and LLM token growth on long calls. Track cost per completed conversation, not just per minute.

### Does Vobiz charge differently for inbound and outbound AI calls?

No. India calling on Vobiz is a flat ₹0.42/min for both inbound and outbound, billed in INR. Numbers in the 140, 1600 and 92 series are ₹1,000/month each.

### Do I need to pay in dollars for an AI voice agent?

For most STT, LLM, TTS and platform vendors, yes, so your bill moves with the exchange rate. Some layers can be paid in rupees, including Vobiz telephony and Sarvam's speech and language models.

## More Related Articles

[![What Latency, Drop Rates, and Audio Failures Are Costing Your Business](https://www.vobiz.ai/blog1706.jpg) Buyer's Guide10 min read ### What Latency, Drop Rates, and Audio Failures Are Costing Your Business How VoIP latency, packet loss and audio failures quietly erode revenue: the thresholds, the numbers, and how telephony quality shapes business outcomes. Read full article ![Ishani Singh](https://www.vobiz.ai/blogpromotion/ishani.png)Ishani Singh 17 June 2026](https://www.vobiz.ai/resources/blogs/buyers-guide/voip-latency-drop-rates-audio-failures/)[![Twilio Alternatives in 2026: Best Voice APIs for India and Voice AI](https://www.vobiz.ai/blog-thumbs/best-twilio-alternatives-india.jpg) Buyer's Guide10 min read ### Twilio Alternatives in 2026: Best Voice APIs for India and Voice AI Twilio alternatives for voice in 2026: Indian numbers, INR billing and AI voice agents. Compare Vobiz, Plivo, Exotel, Telnyx and more, plus a migration guide. Read full article ![Team Vobiz](https://www.vobiz.ai/favicon-192.png)Team Vobiz 19 November 2026](https://www.vobiz.ai/resources/blogs/buyers-guide/best-twilio-alternatives-india/)[![Best Plivo Alternatives in 2026: 6 Voice APIs Compared](https://www.vobiz.ai/blog-thumbs/best-plivo-alternatives.jpg) Buyer's Guide11 min read ### Best Plivo Alternatives in 2026: 6 Voice APIs Compared Plivo alternatives for voice in 2026: Vobiz, Twilio, Exotel, Telnyx, Vonage and Sinch compared, plus how to migrate from Plivo to Vobiz with code. Read full article ![Team Vobiz](https://www.vobiz.ai/favicon-192.png)Team Vobiz 19 November 2026](https://www.vobiz.ai/resources/blogs/buyers-guide/best-plivo-alternatives/)
