Voice AI Agents: The 2026 Business Guide and Pricing

Last Updated: November 10, 2026

Summary: Voice AI agents are software that answers and makes phone calls in natural spoken conversation, combining speech recognition, a large language model and a synthetic voice connected to a phone line. Most business platforms charge per minute: on official pricing pages checked November 10, 2026, Vapi charges $0.05 per minute before model and phone costs, Retell AI $0.055 per minute before model and phone costs, ElevenLabs $0.08 per minute on paid plans and Bland AI $0.12 to $0.14 per minute. In the US, AI-voice calls fall under the TCPA, so outbound calls need prior express consent.

Phone calls are still where a lot of business happens, and voice AI agents are now sold by the minute, like a phone plan. This guide covers how they work, where businesses use them, what the leading platforms cost, what a month of calls adds up to and the legal rules to know first. Every price comes from the vendor's own pricing page or documentation.

Voice AI agents are AI systems that hold a live spoken conversation with a person, usually over the phone, and can take actions during the call, such as checking a calendar, looking up an order or transferring to a human. They differ from old phone menus because the caller speaks naturally instead of pressing 1 or 2.

A voice AI agent is an AI agent whose interface is speech. Like the agents in our explainer on what AI agents are, it follows instructions and calls outside tools, but it must do so while someone waits on the line, so speed matters as much as accuracy.

Voice AI agents work by turning the caller's speech into something a language model can reason about, generating a reply, and speaking that reply back fast enough to feel like a normal conversation. There are two main ways to build that loop.

The chained approach runs three separate stages. Speech-to-text (STT) transcribes what the caller said, a large language model (LLM) decides how to respond and which tools to call, and text-to-speech (TTS) turns the reply into audio. OpenAI's voice agents guide describes the chained pipeline as the choice when you need "control over each speech and text stage," for example to store transcripts or run policy checks before the agent responds, and notes that each component can be replaced independently.

The speech-to-speech approach uses one model that listens to audio and replies in audio directly. OpenAI's same guide describes its Realtime API as a way to "use one model to interpret audio, decide what to do, and respond in speech," with the session handling audio turns, tools, interruptions and handoffs.

Platforms either sell phone numbers directly or connect to a carrier such as Twilio, or to an existing phone system through SIP trunking. Telephony is usually a separate line item: Twilio's US voice pricing page lists $0.0085 per minute for inbound calls to a local number, $0.014 per minute for outbound calls and $1.15 per month for a local number (checked November 10, 2026).

Latency is the pause between the caller finishing a sentence and the agent starting to answer. Long pauses make callers talk over the agent or hang up. OpenAI's guide recommends measuring "how long callers wait for a useful spoken answer," tracking backend time separately and comparing median and 95th percentile values, rather than trusting a single average.

Businesses use voice AI agents mainly for high-volume, repetitive calls where the questions are predictable and the answer lives in a system the agent can reach. Four uses account for most deployments.

Developer voice AI platforms charge between $0.05 and $0.14 per minute in platform fees on the pricing pages below, but several platforms add separate charges for the language model, speech services and phone line. Always compare the all-in cost per minute, not the headline fee.

Prices checked on official vendor pages on November 10, 2026. Prices exclude taxes and change often, so confirm before you sign.

Bland AI's per-minute rate is the closest to all-in on this list, since its pricing page says the rate covers the language model and speech services with no token charges, though telephony is still billed separately. Vapi, Retell AI and ElevenLabs Agents use a modular model: a platform fee plus pass-through costs for the LLM and phone line. Modular pricing lets you pick the model for each piece.

OpenAI bills its Realtime API by token, not by minute. The gpt-realtime-1.5 model page calls it OpenAI's "flagship audio model for voice agents and customer support" and lists audio input at $32 and audio output at $64 per million tokens, with cached audio input at $0.40. Per-minute cost depends on how much each side talks, so test with real calls. Our OpenAI API pricing guide covers the rest of OpenAI's rates.

A business handling 2,000 calls a month at an average of 4 minutes each (8,000 minutes) would pay roughly $469 to $1,259 a month across the setups below, before any separate model charges and, for Google and Bland AI, before phone charges. Here is the math, using only published rates.

Three findings stand out. First, Bland AI's Build plan only beats Start above 14,950 minutes a month, because the $299 fee divided by the $0.02 per-minute saving equals 14,950. Second, ElevenLabs Pro ($99 plus 6,762 overage minutes at $0.08) and Scale ($299 plus 4,262 overage minutes at $0.08) both come to $639.96 for 8,000 minutes, so the choice between those tiers is about concurrency (20 versus 30 calls) rather than price. Third, Google's per-session price favors short calls: at 5 minutes a call works out to $0.10 per minute, but a 2-minute call works out to $0.25 per minute.

Plan limits matter as much as rates. Bland AI's Start plan caps usage at 100 calls a day and 10 concurrent calls, so busy mornings can hit the ceiling even when the monthly average fits.

The leading voice AI agent platforms fall into four groups. Developer platforms built for phone agents (Vapi, Retell AI, Bland AI); voice specialists such as ElevenLabs, whose agents bundle its synthetic voices (see our ElevenLabs pricing guide); model providers such as the OpenAI Realtime API for teams building their own stack; and contact-center suites (Amazon Connect, Google CX Agent Studio) for companies already running on AWS or Google Cloud. Our roundup of AI agent platforms covers non-voice options too.

Voice AI agents are legal for businesses to use, but in the US, calls that use AI-generated voices count as "artificial" voice calls under the Telephone Consumer Protection Act (TCPA), so outbound calls need the called party's prior express consent. This is general information, not legal advice; check with counsel before launching outbound calling.

The US Federal Communications Commission issued a Declaratory Ruling (FCC 24-17, CG Docket No. 23-362) recognizing that calls made with AI-generated voices are "artificial" under the TCPA, according to the FCC's announcement. The FCC's ruling page states that the TCPA's restrictions "encompass current AI technologies that generate human voices," and that such calls require "the prior express consent of the called party." Law firm Kelley Drye's analysis notes the ruling was released February 8, 2024, and that telemarketing calls need prior express written consent, must identify the business responsible for the call and must offer an opt-out.

Most voice AI platforms record and transcribe calls, which brings in wiretapping laws. Federal law, 18 U.S.C. 2511(2)(d), allows recording when one party to the call has given prior consent. Some states require more: California Penal Code 632 prohibits recording a confidential communication "without the consent of all parties." A clear recording notice at the start of every call is the safest default for any business taking calls from multiple states.

The EU AI Act requires disclosure. The European Commission's Article 50 FAQ says AI systems that interact directly with people must be designed so people are informed they are interacting with an AI system, unless that is obvious, and that Article 50 applies from August 2, 2026.

Choose a voice AI agent by starting from one specific call type, then comparing platforms on all-in cost per resolved call, latency on your own integrations, and compliance features.

What is a voice AI agent?
A voice AI agent is software that holds a live spoken conversation, usually by phone, using speech recognition, a large language model and a synthetic voice. It can take actions during the call, such as booking an appointment or transferring to a human.

How much does a voice AI agent cost?
Platform fees on official pricing pages checked November 10, 2026 range from $0.05 per minute (Vapi) to $0.14 per minute (Bland AI Start), and several platforms add separate charges for the language model and phone line. For 8,000 minutes a month, published rates work out to roughly $469 to $1,259 before separate model charges and, for some platforms, phone charges.

Are AI voice agents legal for outbound calls?
Yes, with consent. The FCC ruled in 2024 that AI-generated voices are "artificial" under the TCPA, so outbound AI-voice calls need the called party's prior express consent, and telemarketing calls need prior express written consent.

What is the difference between a voice AI agent and an IVR?
An IVR uses fixed menus where callers press keys or say set words. A voice AI agent understands open-ended speech, answers questions in natural language and can complete tasks, so callers say what they need instead of working through a menu.

Do voice AI agents have to tell callers they are AI?
In the EU, Article 50 of the AI Act requires people to be informed they are interacting with an AI system, unless it is obvious, from August 2, 2026. In the US, the TCPA requires AI-voice calls to identify the business responsible, and disclosing the AI is a sensible default everywhere.

Voice AI agents are now a priced, off-the-shelf business tool: platform fees run from about $0.05 to $0.14 a minute on official pricing pages, with model and phone costs on top for modular platforms. The cheapest headline rate is rarely the cheapest setup, so price the full stack against your real call volume.

Start with one call type, a clear handoff to humans, and consent and disclosure built in from day one.

⭐ Get more AI Business Weekly in Google: add us as a preferred source on Google.

Written by Sameer Khan with AI assistance, fact-checked against the sources linked above.