AI · 10 min read
AI Voice Agent Cost in 2026: Per Minute, Per Month and Development Cost
Per-minute and monthly running costs from Twilio, Deepgram, Vapi, Retell and Bland list prices, with build budgets from $12k to $22k for a pilot agent.
Zain Khalid MalikCTO & Co-founder, Innovation Insight·Published 
Short answer
An AI voice agent costs about $0.05 to $0.10 per minute to run on a self-assembled stack of Twilio, Deepgram, a mini-class model and mid-tier text to speech. A managed platform such as Vapi or Retell runs $0.10 to $0.20 per minute. Realtime speech-to-speech models cost $0.30 or more. All at vendor prices checked October 2026. Building one costs $12k to $22k for a pilot, $22k to $45k for production and $45k to $90k+ for enterprise scope.
AI voice agent cost at a glance
| Cost | Typical range | What sets it |
|---|---|---|
| Running cost per minute, self-assembled stack | $0.05 to $0.10 | Telephony + speech to text + model + text to speech, each at cost |
| Running cost per minute, managed platform | $0.10 to $0.20 | Platform fee plus the same components, often passed through |
| Running cost per minute, realtime speech-to-speech | $0.30 and up | Audio-native models priced per audio token |
| Development: pilot | $12k to $22k | One call flow, one integration, a test number, call review tools |
| Development: production | $22k to $45k | Several flows, CRM or booking integration, transfers, analytics, guardrails |
| Development: enterprise | $45k to $90k+ | Multiple lines and languages, compliance, custom telephony, evaluation at scale |
Build prices are Innovation Insight's published ranges at $25 to $49 an hour. Running costs are derived from the vendors' own pricing pages, all checked October 2026, and they are the part most buyers get wrong, so this guide starts there.
AI voice agent cost per minute: the stack
Every voice agent is four metered services in a loop: the phone line, speech to text, a language model, and text to speech. Managed platforms add a per-minute fee for orchestrating the loop, handling interruptions and keeping latency low. The table lists each component with its published price.
| Component | Vendor and product | Published price | Source, checked October 2026 |
|---|---|---|---|
| Telephony | Twilio US voice | $0.0085 per minute inbound, $0.014 outbound; local number $1.15 a month | twilio.com/en-us/voice/pricing/us |
| Speech to text | Deepgram Nova-3 streaming | $0.0048 per minute promotional ($0.0077 regular) | deepgram.com/pricing |
| Language model | OpenAI, Anthropic or Google via a platform | About $0.008 to $0.045 per call minute depending on model (Vapi's calculator); Retell lists GPT-5.4 mini at $0.024, Gemini 3.0 Flash $0.016, GPT-5.4 $0.08 | vapi.ai/pricing, retellai.com/pricing |
| Text to speech | ElevenLabs, Deepgram Aura-2, Cartesia | ElevenLabs about $0.015 to $0.024 per minute via Vapi; Aura-2 $0.030 per 1,000 characters, roughly $0.01 to $0.025 per call minute depending on how much the agent talks | vapi.ai/pricing, deepgram.com/pricing |
| Platform fee | Vapi | $0.05 per minute plus components at cost; Core $29 a month for 10 concurrent calls, Pro $999 a month minimum for 30; extra concurrency $10 per line; HIPAA add-on $2,000 a month | vapi.ai/pricing |
| Platform fee | Retell AI | Voice engine $0.055 per minute; telephony $0.015 or free with your own SIP; TTS $0.015 to $0.10; all-in $0.07 to $0.31 per minute; 20 concurrent calls free then $8 per line a month; numbers $2 a month | retellai.com/pricing |
| Platform fee | Bland AI | Start $0.14 per minute (10 concurrent, 100 calls a day cap); Build $0.12 per minute plus $299 a month (50 concurrent); rate includes model, STT and TTS; telephony passed through | bland.ai/pricing |
| Platform fee | ElevenLabs Agents | $0.08 per minute within plan minutes, $0.16 beyond; plans from $22 (Creator) to $990 (Business) a month | elevenlabs.io/pricing/api |
| Speech-to-speech model | OpenAI gpt-realtime-2.1 via Retell | $0.38 per minute; gpt-realtime mini $0.07 per minute; OpenAI lists audio at $32 / $64 per 1M tokens for 2.1 and $10 / $20 for mini | retellai.com/pricing, developers.openai.com/api/docs/pricing |
Adding the self-assembled column: Twilio inbound at $0.0085, Deepgram at $0.0048, a mini-class model at about $0.024 and a mid-tier voice at about $0.02 comes to roughly $0.06 per minute, and swapping in a cheaper model and voice gets close to $0.05. Using GPT-5.4 and a premium ElevenLabs voice pushes it past $0.10. On Vapi the same components plus the $0.05 fee land near $0.11; Retell's own example range is $0.07 to $0.31; Bland's bundled rate is $0.12 to $0.14 plus telephony. Speech-to-speech models skip the STT and TTS steps but cost more than all of them combined.
AI voice agent cost per month: three scenarios
Monthly cost is minutes multiplied by the per-minute stack cost, plus fixed fees for phone numbers, concurrency and any plan minimum. The scenarios below apply the per-minute ranges above; fixed fees are shown separately because they depend on how many calls happen at the same time, not on volume.
| Minutes a month | Self-assembled stack ($0.05 to $0.10) | Managed platform ($0.10 to $0.20) | Speech-to-speech ($0.30+) | Fixed fees to add |
|---|---|---|---|---|
| 1,000 (a small clinic or agency line) | $50 to $100 | $100 to $200 | $300+ | Numbers ($1.15 to $2 each); Vapi Core $29 if you need 10 concurrent calls |
| 10,000 (a busy support line) | $500 to $1,000 | $1,000 to $2,000 | $3,000+ | Extra concurrency at $8 to $10 per line a month beyond the free lines |
| 50,000 (a contact centre queue) | $2,500 to $5,000 | $5,000 to $10,000 | $15,000+ | Bland Build $299 a month or Vapi Pro $999 minimum; volume pricing on request from every vendor |
Vapi's own example for 1,000 minutes a month is $82 to $129, which sits inside the managed-platform band. At 50,000 minutes the difference between a self-assembled stack and a managed one is several thousand dollars a month, which is the point where owning the orchestration layer starts to pay for itself. Below that, a platform usually wins on time to launch and latency engineering you would otherwise do yourself.
AI voice agent development cost
Running cost is only half the budget. A voice agent needs call-flow design, the integrations that make it useful (booking, CRM, order lookups), transfer rules, a test suite of recorded calls and the review tools your team will use to listen and tune. Our calculator prices the voice agent feature at 160 hours and a full inbound-call agent with document answers, admin and two integrations at $23k to $42k over 10 to 18 weeks; our published tiers below bracket that.
| Tier | What is included | Cost | Timeline |
|---|---|---|---|
| Pilot | One call flow on one number, one integration, interruption handling, call recording and review, 50-call evaluation set | $12k to $22k | 3 to 5 weeks |
| Production | Several flows, CRM or scheduling integration, warm transfer to staff, voicemail, analytics, guardrails, multilingual if needed | $22k to $45k | 6 to 10 weeks |
| Enterprise | Multiple lines and languages, your own SIP trunk, compliance (HIPAA or PCI scope), load testing, evaluation at scale, SLAs | $45k to $90k+ | 10 to 16 weeks |
| Component | Hours | Approximate cost |
|---|---|---|
| Web app: setup, architecture, release | 140 | $3.9k to $5.3k |
| Design: clean and simple | 40 | $1.1k to $1.5k |
| Admin panel | 70 | $2k to $2.7k |
| Dashboards and reports | 48 | $1.3k to $1.8k |
| AI answers from your documents (RAG) | 180 | $5k to $6.8k |
| AI voice agent (phone or in-app) | 160 | $4.5k to $6.1k |
| 2 third-party integrations | 72 | $2k to $2.7k |
| Project management, QA and release (35%) | 249 | $7k to $9.5k |
What drives voice agent cost
- Latency budget. Callers notice anything over about a second of silence, so model choice, streaming TTS and where the servers run are engineering decisions with cost consequences.
- Integrations. A booking agent that cannot book is a demo. Each system the agent reads or writes is a week of work plus tests.
- Transfers and fallbacks. Warm transfer with context, voicemail capture and what happens when a vendor is down.
- Compliance. Call recording consent, PCI scope if callers read card numbers, HIPAA for health lines (Vapi charges $2,000 a month for its HIPAA add-on).
- Languages and accents. Each language needs its own evaluation calls and often a different voice and STT model.
- Evaluation. A library of recorded test calls run on every change is what keeps the agent from regressing when a model is updated.
Choosing a platform
Vapi, Retell and Bland take different approaches to the same problem: Vapi passes components through at cost plus a per-minute fee, Retell bundles its voice engine and lets you pick each component, and Bland sells an all-in per-minute rate. The Vapi vs Retell vs Bland comparison covers latency, telephony options and when to build on raw APIs instead. For a text-first assistant, the AI chatbot development cost guide is the right starting point.
How to reduce AI voice agent cost
- Measure your real call lengths before choosing a stack; average handle time sets everything.
- Use a mini-class model for most turns and escalate only when the conversation needs reasoning.
- Bring your own SIP trunk at volume; Retell waives its telephony fee when you do.
- Cap call duration and detect voicemail early so you do not pay for minutes nobody hears.
- Pick the cheapest voice that your callers rate as natural in a blind test; premium voices are often several times the price.
- Negotiate volume pricing once you pass 10,000 minutes a month; every vendor offers it.
See our voice AI agent development service for how we build and run these agents.
Sources
- Innovation Insight voice AI agent development, AI agent cost calculator and Vapi vs Retell vs Bland
- Innovation Insight AI chatbot development cost
- Vapi pricing: https://vapi.ai/pricing
- Retell AI pricing: https://www.retellai.com/pricing
- Bland AI pricing: https://www.bland.ai/pricing
- ElevenLabs API pricing: https://elevenlabs.io/pricing/api
- Deepgram pricing: https://deepgram.com/pricing
- Twilio US voice pricing: https://www.twilio.com/en-us/voice/pricing/us
- OpenAI API pricing: https://developers.openai.com/api/docs/pricing
Need a number for your project?
Send a short brief and get a written estimate.
A senior engineer replies within one business day. No sales call required.
Zain Khalid Malik
CTO & Co-founder, Innovation Insight
Zain owns architecture, engineering standards and the platform team at Innovation Insight. He sets the bar for code quality, security and the tooling every squad ships with.
LinkedInRelated questions.
How much does an AI voice agent cost per minute?
About $0.05 to $0.10 per minute on a self-assembled stack, $0.10 to $0.20 on a managed platform such as Vapi, Retell or Bland, and $0.30 or more with realtime speech-to-speech models, based on vendor prices checked October 2026.
How much does an AI voice agent cost per month?
1,000 minutes costs roughly $50 to $200 depending on the stack; 10,000 minutes $500 to $2,000; 50,000 minutes $2,500 to $10,000, plus phone numbers, concurrency lines and any plan minimum.
How much does it cost to develop an AI voice agent?
$12k to $22k for a pilot, $22k to $45k for a production agent with integrations and transfers, and $45k to $90k+ for enterprise scope with a senior offshore team.
Is there an AI voice agent cost calculator?
Yes. Our AI agent cost calculator estimates the monthly bill from your call minutes and chosen stack, and the app cost calculator prices the build.
Vapi, Retell or Bland: which is cheapest?
At low volume they are close: Vapi is $0.05 a minute plus components, Retell's own range is $0.07 to $0.31 all-in, Bland is $0.12 to $0.14 bundled. Component choices (model and voice) move the price more than the platform does.
Can a voice agent be HIPAA compliant?
Yes, with a platform that signs a BAA (Vapi sells a HIPAA add-on at $2,000 a month) or a self-assembled stack on HIPAA-eligible services, plus recording consent and access controls on transcripts.