AI Voice Agent Cost: 2026 Pricing Breakdown & Calculator

An AI voice agent costs between $0.05 and $0.50 per minute of conversation in published pricing as of October 2026, depending on the vendor, call volume and which components are bundled. The total monthly cost depends on call volume, average call length and whether you pay for platform fees, telephony, speech services and language-model inference separately or in one rate.

This breakdown covers every cost component, the pricing models vendors use, a formula you can apply to your own call volumes, and how to keep costs predictable. All prices were checked in October 2026 and link to the vendor's public pricing page or documentation.

Key takeaways

1. Bland AI charges $0.09 per minute for outbound calls and $0.12 per minute for inbound calls as of October 2026, according to its pricing page, with telephony, speech-to-text, language model and text-to-speech included in the per-minute rate.
2. Vapi charges $0.05 per minute for the platform orchestration fee as of October 2026, according to its pricing page, plus separate costs for speech-to-text, language model, text-to-speech and telephony that you bring or buy through Vapi's integrated providers.
3. Retell AI charges $0.08 per minute for its platform fee as of October 2026, according to its pricing page, with speech-to-text and text-to-speech included, but language-model costs and telephony billed separately.
4. Telephony costs $0.004 to $0.02 per minute depending on the country and provider, according to Twilio's pricing page as of October 2026, and phone-number rental adds $1 to $15 per month per number.
5. Language-model inference can cost $0.01 to $0.15 per minute of conversation depending on the model, input and output token counts, and provider pricing, based on OpenAI and Anthropic pricing as of October 2026.
6. A 1,000-call-per-month pilot at 3 minutes average call length costs roughly $135 to $1,500 per month depending on the vendor, pricing model and whether you use a bundled rate or pay for each component separately, using illustrative assumptions and October 2026 published rates.

How much does an AI voice agent cost?

AI voice agents are priced per minute of conversation, per seat or agent, per outcome, or under enterprise contracts with volume commitments. Published per-minute rates range from $0.05 to $0.50 as of October 2026, but the all-in cost depends on whether telephony, speech services and language-model inference are bundled or billed separately.

Most vendors charge for connected call time only. Ringing and failed attempts do not count.

Pricing checked October 2026. Each price links to the vendor's pricing page.

The bundled per-minute model is the most common for sales and lead-qualification agents. The unbundled model gives you control over which speech and language-model providers you use, but requires you to track costs across multiple vendors. Subscription pricing works for teams that want a fixed monthly cost and predictable usage.

Enterprise contracts are for contact centers running thousands of calls per day.

What are the cost components?

Every AI voice agent incurs costs for orchestration, speech recognition, language understanding, speech synthesis and telephony. Some vendors bundle all components into one per-minute rate. Others charge separately for each.

Understanding the components helps you compare pricing and estimate your total cost.

Platform or orchestration fee

The platform fee covers the software that coordinates the call: routing audio streams, managing turn-taking, handling interruptions, logging transcripts and integrating with your CRM or calendar. Vapi charges $0.05 per minute for orchestration as of October 2026, according to its pricing page. Retell AI charges $0.08 per minute for the platform with speech-to-text and text-to-speech included, according to its pricing page as of October 2026.

Bland AI bundles the orchestration fee into its all-in per-minute rate.

Open-source frameworks such as Pipecat and LiveKit charge no platform fee, but you pay for the compute to run the orchestration logic and the engineering time to build and maintain it.

Speech-to-text (STT)

Speech-to-text converts the caller's voice into text that the language model can process. Deepgram charges $0.0043 per minute for its Nova-2 model as of October 2026, according to its pricing page. OpenAI's Whisper API costs $0.006 per minute as of October 2026, according to OpenAI's pricing page. AssemblyAI charges $0.00025 per second, which is $0.015 per minute, for real-time transcription as of October 2026, according to its pricing page.

Some vendors bundle STT into the platform fee. Others let you bring your own provider or charge separately.

Latency matters more than cost for real-time voice. Deepgram's Nova-2 and AssemblyAI's real-time models are optimized for low latency. Whisper is accurate but slower. Most production voice agents use Deepgram or AssemblyAI.

Language model (LLM)

The language model generates the agent's responses. Cost depends on the model, the number of input tokens (the conversation history and system prompt) and the number of output tokens (the agent's reply).

OpenAI's GPT-4o costs $2.50 per million input tokens and $10.00 per million output tokens as of October 2026, according to OpenAI's pricing page. Anthropic's Claude 3.5 Sonnet costs $3.00 per million input tokens and $15.00 per million output tokens as of October 2026, according to Anthropic's pricing page. GPT-4o-mini costs $0.150 per million input tokens and $0.600 per million output tokens as of October 2026, according to OpenAI's pricing page.

A 3-minute call with a 2,000-token system prompt and 1,500 tokens of conversation history might generate 4,000 input tokens and 800 output tokens per turn, with 6 to 10 turns. That is roughly 30,000 input tokens and 6,000 output tokens per call. At GPT-4o pricing, that call costs $0.075 for input and $0.060 for output, or $0.135 total. At GPT-4o-mini pricing, the same call costs $0.0045 for input and $0.0036 for output, or $0.0081 total.

The difference between a frontier model and a small model can be $0.01 to $0.15 per minute of conversation.

Vendors that bundle the language model into the per-minute rate usually use GPT-4o-mini or a similar small model. If you need GPT-4o or Claude 3.5 Sonnet for complex reasoning, you pay separately or the bundled rate is higher.

Text-to-speech (TTS)

Text-to-speech converts the language model's text response into audio. ElevenLabs charges $0.18 per 1,000 characters for its Turbo v2.5 model as of October 2026, according to its pricing page. OpenAI's TTS API costs $15.00 per million characters, which is $0.015 per 1,000 characters, for the standard voice as of October 2026, according to OpenAI's pricing page. Deepgram's Aura model costs $0.015 per 1,000 characters as of October 2026, according to its pricing page.

A 3-minute call with 6 agent turns of 100 characters each is 600 characters, or roughly $0.001 to $0.11 per call depending on the provider and voice quality.

ElevenLabs voices are the most natural but cost 10 to 12 times more than OpenAI or Deepgram. Most production agents use Deepgram Aura or OpenAI for cost and latency. ElevenLabs is used when voice quality is a differentiator, such as customer service for premium brands.

Telephony and phone numbers

Telephony is the cost of connecting the call to the public telephone network. Twilio charges $0.0085 per minute for outbound calls to US numbers and $0.0085 per minute for inbound calls as of October 2026, according to Twilio's pricing page. Calls to mobile numbers in other countries cost $0.01 to $0.20 per minute depending on the destination, according to Twilio's pricing page.

Vonage charges similar rates. Some vendors include telephony in the per-minute rate. Others charge separately or let you bring your own Twilio or Vonage account.

Phone-number rental costs $1.00 per month for a US local number and $2.00 per month for a US toll-free number as of October 2026, according to Twilio's pricing page. International numbers cost $1.00 to $15.00 per month depending on the country. If you run agents in 10 countries, phone-number rental alone can be $50 to $150 per month before any calls.

Concurrency or capacity

Concurrency is the number of simultaneous calls the agent can handle. Some vendors charge per concurrent call or per seat. Synthflow charges $99 per month for one agent with 500 included minutes as of October 2026, according to its pricing page. If you need 10 concurrent agents to handle peak call volume, you pay $990 per month before usage.

Other vendors charge per minute with no concurrency limit, but you pay for the compute and API capacity to handle the peak. Cloud infrastructure to run 10 concurrent calls costs roughly $50 to $200 per month depending on the region and instance type, based on AWS EC2 pricing as of October 2026.

Implementation and integration work

Implementation includes writing the system prompt, designing the conversation flow, connecting the agent to your CRM, calendar or helpdesk, and testing edge cases. Most vendors do not publish implementation fees. Agencies and consultancies charge $5,000 to $50,000 for a production-ready agent depending on complexity, according to third-party sources.

If you build in-house, budget 40 to 200 engineering hours for the first agent. Subsequent agents reuse the integration work and take fewer hours.

Ongoing QA and conversation design

Voice agents require continuous quality assurance: listening to call recordings, reading transcripts, updating the prompt when the agent mishandles a scenario, and retraining on new edge cases. Budget 5 to 20 hours per week for a team running 1,000 to 10,000 calls per month, based on third-party sources. At $50 to $150 per hour for a conversation designer or QA analyst, that is $1,000 to $12,000 per month in labour.

Some vendors offer managed QA as a professional service, but pricing is not publicly listed.

Compliance work

If you call consumers in the US, you must comply with the Telephone Consumer Protection Act (TCPA), which requires prior express written consent for autodialed or prerecorded calls to mobile numbers, according to the Federal Communications Commission. If you record calls, you must comply with state recording-consent laws, which vary by state. If you handle health information, HIPAA applies. If you process payment card data, PCI DSS applies. If you call or store data on EU residents, GDPR applies.

Compliance work includes legal review of your consent flow, call scripts and data-retention policies, and technical controls such as encryption, access logs and data-deletion workflows. Budget $10,000 to $100,000 for legal and technical compliance work before launch, and ongoing audit and monitoring costs, based on third-party sources.

Recommend legal review for any deployment that handles regulated data or calls consumers.

How do you estimate your monthly cost?

Use this formula to estimate your monthly cost:

monthly cost ≈ (calls per month × average minutes per call × all-in per-minute rate) + platform fees + phone-number fees + add-ons

Each input depends on your use case and call patterns. Calls per month is the number of connected calls, not dials. If you dial 1,000 numbers and 300 connect, you pay for 300 calls. Average minutes per call is the conversation length, not including ringing or hold time. Most vendors charge only for connected time.

The all-in per-minute rate is either the vendor's bundled rate or the sum of platform, STT, LLM, TTS and telephony costs. Platform fees include any monthly subscription or seat charges. Phone-number fees are the rental cost for local, toll-free or international numbers. Add-ons include CRM integrations, analytics dashboards and professional services.

Call length varies by use case. Inbound lead qualification averages 2 to 4 minutes. Outbound appointment setting averages 1 to 3 minutes when the contact answers. Inbound customer support averages 4 to 8 minutes. Outbound surveys average 3 to 6 minutes.

Longer calls cost more per call but often deliver higher value per conversation.

Connect rate affects total cost. If you dial 1,000 numbers and 30% connect, you make 300 calls. If you retry no-answers, you might dial 1,500 times to get 300 connects. Most vendors charge only for connected calls, so retries do not add per-minute cost, but they do consume concurrency and telephony capacity.

Transfers and voicemail change the calculation. If the agent transfers 20% of calls to a human, you pay for the agent's portion of the call, not the human's. If the agent leaves a voicemail, some vendors charge for the message length, others do not. Check the vendor's pricing page for transfer and voicemail policies.

Illustrative scenarios as of October 2026. Per-minute rates link to vendor pricing pages.

The unbundled scenario assumes Vapi's $0.05 per minute platform fee, Deepgram Nova-2 at $0.0043 per minute, GPT-4o-mini at roughly $0.027 per minute for a 3-minute call with illustrative token counts, Deepgram Aura at roughly $0.002 per minute, and Twilio telephony at $0.0085 per minute. The subscription scenario assumes Synthflow's $99 per month per agent with 500 included minutes and an illustrative overage rate of $0.108 per minute, which is not publicly listed. The high-volume scenario assumes an illustrative volume discount to $0.07 per minute, which some vendors offer but do not publish.

All numbers are illustrative assumptions to show the formula, not typical results.

What does it cost compared with the alternatives?

The alternative to an AI voice agent is a human making or taking the call, an answering service, or an IVR menu. Human labour costs $15 to $50 per hour depending on the role, location and whether the worker is in-house or outsourced, according to the US Bureau of Labor Statistics as of 2025. A sales development representative in the US earns a median $25 per hour, according to BLS data. An outsourced call-center agent costs $10 to $20 per hour, according to third-party sources.

A 3-minute call handled by a $25-per-hour SDR costs roughly $1.25 in labour. The same call handled by an AI voice agent at $0.09 per minute costs $0.27, or roughly 22% of the human cost.

The cost advantage grows with volume. A team making 10,000 calls per month at 3 minutes per call is 30,000 minutes, or 500 hours. At $25 per hour, that is $12,500 in labour. At $0.09 per minute, the AI agent costs $2,700, or 22% of the labour cost. At 50,000 calls per month, the labour cost is $62,500 and the AI cost is $10,500, assuming an illustrative volume discount to $0.07 per minute.

Answering services charge $0.75 to $1.50 per call or $0.80 to $1.20 per minute, according to third-party sources. A 3-minute inbound call costs $2.25 to $4.50 with an answering service and $0.27 to $0.36 with an AI voice agent at bundled pricing. The AI agent is 6% to 12% of the answering-service cost.

IVR systems cost $50 to $500 per month for the software plus telephony costs, according to third-party sources. An IVR handles simple routing but cannot have a conversation. An AI voice agent costs more per call than an IVR but handles qualification, objection handling and appointment booking that an IVR cannot.

The comparison depends on whether the call requires a conversation or just a menu.

The labour-cost comparison assumes the AI agent handles the call end-to-end. If the agent qualifies the lead and transfers to a human closer, you pay for both the agent's time and the human's time. If the agent books 30% of calls and transfers 70%, the blended cost is lower than a human handling 100% of calls but higher than the agent alone.

What hidden costs should you budget for?

Subscription plans include a block of minutes. Overage rates are often 20% to 50% higher than the included rate, based on third-party sources. If you exceed your plan regularly, you pay more per minute than the headline rate.

If the agent retries no-answers or busy signals, you pay for concurrency and telephony capacity even if the call does not connect. Some vendors charge a small fee for failed attempts. Check the pricing page.

If the agent writes to Salesforce, HubSpot or another CRM, you may hit API rate limits or pay for additional API calls. Salesforce charges for API calls above the included limit in some plans, according to Salesforce documentation.

Call recordings and transcripts consume storage. If you store 10,000 calls per month at 3 minutes each with audio and text, that is roughly 30 GB of audio and 5 GB of text per month. Storage costs $0.023 per GB per month on AWS S3 as of October 2026, or roughly $0.80 per month. Retention policies that keep data for years add up.

If you operate in a regulated industry, you may need third-party audits of your call scripts, consent flows and data handling. Budget $5,000 to $50,000 per year for compliance audits, based on third-party sources.

If you fine-tune a language model on your data or commission a custom voice clone, vendors charge setup fees and higher per-use rates. ElevenLabs charges for voice cloning, according to its pricing page, but the fee is not publicly listed.

Calls to mobile numbers in some countries cost $0.10 to $0.50 per minute, according to Twilio's pricing page. If you call internationally, check the per-country rates.

How do you keep costs predictable?

Can you commit to a volume tier?

Most vendors offer volume discounts for annual commitments. If you commit to 100,000 minutes per year, the per-minute rate may drop from $0.09 to $0.07 or lower, based on third-party sources. The discount is not usually published. You negotiate it with the vendor.

The risk is paying for unused minutes if your volume is lower than expected. Commit to a volume tier only if you have historical call data or a pilot that validates the forecast.

Does the vendor cap overages?

Some subscription plans cap overage charges at a maximum monthly amount. If your plan includes 500 minutes and you use 2,000 minutes, a cap limits the overage cost to a fixed amount instead of charging the overage rate for all 1,500 excess minutes.

Not all vendors publish overage caps. Ask during the sales process.

Should you set concurrency limits?

If you pay per minute with no concurrency limit, a sudden spike in call volume can drive up costs. Setting a concurrency limit in your telephony provider (for example, 10 simultaneous calls) prevents runaway costs but means some calls go to voicemail or get a busy signal during the spike. The trade-off is cost control versus availability.

For pilots, set a low concurrency limit and raise it as you validate ROI.

Can you control call length?

Shorter calls cost less. If the agent can qualify a lead in 2 minutes instead of 4, you cut the per-call cost in half. Design the conversation to ask disqualifying questions early. If the contact is not a fit, end the call politely.

Some vendors let you set a maximum call length in the agent configuration. If the call hits the limit, the agent wraps up and ends the call.

Does model choice affect the rate?

If you use an unbundled vendor, switching from GPT-4o to GPT-4o-mini can cut language-model costs by 90%. The trade-off is reasoning quality. GPT-4o-mini handles simple scripts well but struggles with complex objection handling or multi-step logic.

Test both models on your use case. If GPT-4o-mini performs well enough, use it and save $0.10 per minute.

Should you pilot before an annual contract?

Run a 1,000 to 5,000 call pilot on a month-to-month plan before committing to an annual contract. Measure connect rate, average call length, conversion rate and cost per outcome. If the pilot ROI is positive, negotiate an annual contract with a volume discount. If the ROI is unclear, extend the pilot or try a different use case.

Do not lock in annual spend until you have data.

Frequently asked questions

How much does an AI voice agent cost per minute?

AI voice agents cost $0.05 to $0.50 per minute of conversation as of October 2026, depending on the vendor and whether telephony, speech-to-text, language model and text-to-speech are bundled or billed separately. Bland AI charges $0.09 per minute for outbound calls with all components included, according to its pricing page as of October 2026. Vapi charges $0.05 per minute for the platform fee plus separate costs for speech, model and telephony, according to its pricing page.

The all-in cost depends on which providers you use and your call volume.

Is telephony included in the per-minute rate?

Some vendors include telephony in the per-minute rate. Others charge separately or let you bring your own Twilio or Vonage account. Bland AI includes telephony in its $0.09 per minute outbound rate, according to its pricing page as of October 2026. Vapi charges separately for telephony, according to its pricing page.

Check the vendor's pricing page to see what is included. If telephony is separate, add $0.004 to $0.02 per minute depending on the destination country.

What is the cheapest way to start?

The cheapest way to start is a pay-as-you-go plan with no monthly minimum and a bundled per-minute rate. Bland AI's $0.09 per minute outbound rate has no monthly fee and includes all components, according to its pricing page as of October 2026. A 100-call pilot at 3 minutes per call costs $27 in usage plus $1 for a phone number, or $28 total.

Open-source frameworks such as Pipecat have no platform fee, but you pay for compute, speech, model and telephony separately, and you need engineering time to build and deploy the agent.

Are there setup fees?

Most self-service vendors do not charge setup fees. Enterprise vendors and agencies charge $5,000 to $50,000 for implementation, integration and conversation design, according to third-party sources. If you build in-house on a self-service platform, there is no setup fee, but you pay for engineering time.

If you buy a managed service or white-glove onboarding, ask the vendor for the setup fee during the sales process. Setup fees are not usually published.

Do you pay for failed or abandoned calls?

Most vendors charge only for connected calls, not for ringing, busy signals or no-answers. If the agent dials a number and the contact does not pick up, you do not pay the per-minute rate, but you may pay a small telephony fee for the attempt. Twilio charges $0.0085 per minute for connected calls and does not charge for failed attempts, according to its pricing page as of October 2026.

If the contact picks up and hangs up immediately, you pay for the connected time, even if it is only a few seconds.

How much does it cost to run your own voice agent on open-source infrastructure?

Running your own voice agent on open-source infrastructure such as Pipecat or LiveKit costs $0.03 to $0.15 per minute depending on which speech, model and telephony providers you use, plus compute costs. A 3-minute call using Deepgram STT at $0.0043 per minute, GPT-4o-mini at roughly $0.027 per minute, Deepgram TTS at roughly $0.002 per minute, and Twilio telephony at $0.0085 per minute costs roughly $0.126 in API fees. Compute to run the orchestration logic costs $50 to $200 per month for 10 concurrent calls on AWS, according to AWS EC2 pricing as of October 2026.

You also pay for engineering time to build, deploy and maintain the agent.

Does the price include CRM integration?

Some vendors include CRM integration in the platform fee. Others charge separately or require you to build the integration yourself. Synthflow includes HubSpot and Salesforce integrations in its $99 per month plan, according to its pricing page as of October 2026. Vapi provides webhooks and API access but does not include pre-built CRM connectors, according to its documentation.

If the vendor does not include the integration, budget 10 to 40 engineering hours to build and test the connection to your CRM.

Can you negotiate a lower rate?

Most vendors negotiate volume discounts for annual commitments or high call volumes. If you commit to 100,000 minutes per year or more, the per-minute rate may drop by 10% to 30%, based on third-party sources. Discounts are not usually published. Contact the vendor's sales team with your projected volume and ask for a quote.

Month-to-month plans at published rates are usually non-negotiable.

Last updated: October 2026.