AgentVoice vs Retell

Retell AI publishes a rate for every part of a voice agent, so you can work out what an agent will cost before anyone builds it. AgentVoice builds on Retell AI for clients who want a production phone agent without assembling and maintaining the stack themselves. The Retell AI pricing below comes from Retell’s pricing page as of September 2026, followed by the work we do on Retell and the cases where we’d build somewhere else.

Retell AI pricing, line by line

Retell charges pay-as-you-go with no platform subscription, and it quotes voice agents at $0.07 to $0.31 a minute. That range is the sum of separate rates, since every minute carries Retell’s $0.055 Voice Infra fee plus a voice, a language model and telephony, each priced on its own.

The low end is a Retell voice at $0.015 a minute, GPT 5 nano at $0.0016 and your own SIP trunk, which Retell doesn’t charge for, and that comes to about $0.072 a minute. An agent with an ElevenLabs voice at $0.040, Claude 5 Sonnet at $0.064 and US telephony at $0.015 comes to $0.174 a minute before add-ons. The table has the rest of the rates from Retell’s pricing page.

New accounts start with $10 in free credits. Enterprise pricing is custom and adds dedicated implementation support, custom MSA and DPA terms, role-based access control, custom SSO and 24/7 support through a dedicated portal.

The billing rules behind Retell AI pricing

Retell bills per second, doesn’t charge for failed calls, and stops the AI voice agent fee once a call is transferred to a person. Silence is billed, though, and calls that open with a dynamic message carry a 10-second minimum.

The rule we design around is prompt length. When an agent’s prompt passes 4,000 tokens, Retell multiplies the billed duration by the token count divided by 4,000. An 8,000-token prompt bills a three-minute call as six minutes.

Concurrency is the other line to plan for. Every account includes 20 concurrent calls, and each additional line costs $8 a month. Calls above the limit can still connect as burst calls, but Retell adds $0.10 a minute to the entire call, and burst capacity tops out at three times your limit or your limit plus 300, whichever is lower.

Who Retell AI suits

Retell describes itself as a platform to “build, test, deploy, and monitor AI voice and chat agents,” and it’s set up for business teams and developers alike. You can build entirely in the dashboard with no code, or go fully programmatic through the API, SDKs and an MCP server. Its Conversation Flow editor is a drag-and-drop, node-based builder meant for “structured, high-stakes calls,” and Single Prompt agents handle open-ended conversations.

Around those sit an AI copilot called Conductor that helps build and test agents, and Retell Workflows, launched August 24, 2026, for the steps that happen before a call connects and after it ends. Testing tools include an LLM playground, simulation testing, batch tests and A/B tests.

That makes Retell a good fit for a US business with a defined call process and a staff member who’ll own the flows, and for smaller healthcare practices, because Retell’s BAA is self-signed at “no additional fee” on pay-as-you-go. It fits less well when data location matters, since Retell says it doesn’t “currently operate services within the European Union,” documents no self-hosted option, and keeps call data forever by default unless you set a retention period between 1 and 730 days per agent.

Our Retell AI builds

We usually start from Conversation Flow when a call has required steps, like an intake that has to collect the same details every time, and from Single Prompt when the conversation is looser. Then we connect the agent to the systems behind the call and test it against the calls your team handles every day. The work includes:

• Screening spam calls and identifying callers before they reach your staff
• Answering questions from a Retell knowledge base built from your site and documents, or synced from Google Drive, OneDrive or Notion
• Updating your CRM and tagging leads, with Retell's HubSpot and Salesforce tools or custom functions for other systems
• Booking appointments through Cal.com or your own scheduler and texting a confirmation
• Warm transfers to your team when a caller needs a person
• Follow-up actions and a call report after each call, through Retell Workflows or your own webhooks

For outbound work, Retell runs batch calls from a CSV with scheduling, detects voicemail and IVR menus, and can press keypad digits to get through a phone tree. We also build with the billing rules in mind. Keeping prompts under 4,000 tokens, moving reference material into the knowledge base at $0.005 a minute, and choosing the voice and model for each agent keep the per-minute cost predictable.

Retell usage is billed to your own Retell account. We quote the build separately after scoping the calls with you, and we keep tuning the agent after launch.

When we'd build somewhere else

We check the EU question first. Because Retell doesn’t operate there, callers or data that must stay in the EU point us to Vapi, which runs a separate, isolated EU region, or to Synthflow’s EU cluster for teams ready for a Synthflow enterprise contract. Our head-to-head of Vapi and Retell goes into more detail.

Ownership comes next. Retell documents no self-hosted deployment, and its nearest option is a dedicated server add-on on Enterprise. ElevenLabs offers private deployments inside your AWS or Google Cloud account, Bland offers VPC and on-premise deployment on its Enterprise plan, and we can deploy our engine with the source code into your Azure or AWS account or onto your own servers, with no per-minute platform fee.

Voice billing can also tip the decision. Retell hosts all seven of its voice providers and doesn’t document using your own voice or transcription keys, so an ElevenLabs voice bills at Retell’s $0.040 rate. On Vapi you can connect your own ElevenLabs key, and when we build on ElevenLabs itself, the voice is part of ElevenLabs’ own per-minute pricing.