Retell AI Pricing Explained: Voice Engine, Models, Phone, Add-Ons

Retell AI costs $0.07 to $0.31 a minute on pay-as-you-go, built from four parts: a fixed $0.055 voice infra fee (Retell lists no separate speech-to-text charge), the AI model ($0.0016 to $0.32), the voice ($0.015 for most, up to $0.10 for ElevenLabs v3) and a US Retell number ($0.015). Add-ons cost extra, and Enterprise is priced by quote.

Key Takeaways

• Retell's $0.055 voice engine comes first; the model, voice and phone cost extra.
• Pay-as-you-go runs on prepaid credits, and calls stop when the balance reaches $0.
• Retell charges ElevenLabs voices for every call minute; on your own key, you pay only for what the agent says.

This post is part of our guide to Retell AI alternatives. That guide weighs the platforms against each other. This one stays with Retell and works out what a month on it costs, using Retell's own published prices, with each sum shown.

The four charges on every Retell minute

Every minute of a Retell call is billed in four parts, and Retell's pricing page lists the price of each one.

The first is the voice engine, which Retell calls Retell Voice Infra, at $0.055 a minute. The pricing page says "No platform fees", while Retell's billing guide calls this line "the flat per-minute platform cost of running the real-time call". Retell lists no separate speech-to-text charge, so the $0.055 appears to cover transcription too.

The second is the AI model. Retell bills the model itself, per minute. Prices run from $0.0016 for GPT 5 nano to $0.32 for GPT 6 Astra; GPT 5.4 mini is $0.024 and GPT 4.1 is $0.045. The Fast Tier, on dedicated servers, costs up to twice as much. Retell also sells OpenAI's realtime speech-to-speech models at $0.07 to $0.38 a minute.

The third is the voice. Retell's own voices and most others cost $0.015 a minute. ElevenLabs voices cost more, from $0.040 for Flash up to $0.10 for v3 and v4. Each is a flat rate per call minute, however much the agent talks.

The fourth is the phone line. Calls on a Retell number cost $0.015 a minute in the US, $0.10 in the UK and up to $0.80 in the Philippines. Connect your own carrier over SIP and Retell charges nothing for the line; your carrier bills you instead.

Retell's calculator shows $0.11 a minute for its default setup of GPT 5.1 and Retell's voices on your own carrier.

Retell's two plans and the extras

Retell has two plans. Pay-as-you-go uses the public rates above, while Enterprise is priced by quote and adds dedicated support and contract options.

The business associate agreement (BAA) that HIPAA requires comes at no additional fee on either plan. Retell's security overview also lists on-prem. Retell offers it through Enterprise sales; it is closed source and publishes no price or install details for it.

Add-ons are charged on the call minutes that use them. A knowledge base adds $0.005 a minute, and so do noise removal and safety guardrails, each. Removing personal details adds $0.01 a minute. AI quality checks score a sample of your calls, free for the first 100 minutes and then $0.10 a minute. Outbound extras are per call: batch calling adds $0.005 a dial, and branded caller ID adds $0.10 an outbound call.

Monthly items go on your card at the end of each billing cycle, outside the credit balance. A Retell number costs $2 a month, a US toll-free number $5 (plus $0.06 a minute for inbound calls), and a verified number $10. Each concurrent call above 20 costs $8, each knowledge base above the first 10 costs $8, and SMS costs $20 a month plus $0.01 a message.

Our Vapi vs Retell AI comparison sets these terms beside Vapi's, and Vapi Pricing Explained breaks down Vapi's bill the same way.

Open Source Alternative to Vapi / Retell

Self-hosted voice agent platform — no per-minute fees

dograh-hq/dograh

Star on GitHub

Billing rules to know before you budget

A few billing rules decide how many minutes you pay for, and the prompt-length rule can raise a bill at the same call volume.

Pay-as-you-go workspaces pay from prepaid credits. Usage comes off the balance as it happens, and at $0, calls stop until you buy credits or auto recharge tops you up. Credits never expire, but they are not refundable. Enterprise stays on monthly invoices.

Calls are billed to the second, and calls that fail to connect cost nothing. Silence and hold time are billed, because speech-to-text keeps listening. After a transfer, the agent charges stop and only the phone line keeps billing. Calls shorter than 10 seconds that open with an AI-written greeting are billed as 10 seconds.

Prompt length is the rule to watch. Once the text Retell sends to the model passes 4,000 tokens, the billed duration is multiplied by tokens ÷ 4,000. The count includes the global prompt, tool descriptions, the current step's prompt, the conversation so far, tool results and any knowledge base text pulled in. In Retell's own example, 4,800 tokens turns a 60-second call into 72 billed seconds. The conversation grows with every turn, so long calls on big prompts are where this shows up.

Testing is billed at the same rates as live calls, with no free test tier. If you switch on concurrency burst, calls above your limit still go through, up to three times your limit or your limit plus 300, whichever is lower. Each burst call carries $0.10 a minute extra for its whole length.

What Retell costs for three example businesses

Retell has no single price, so here are three example businesses with every line added up from Retell's list prices. Each uses Retell's own US numbers and leaves out taxes and the $10 trial credit.

A small support line

This is a small inbound line: 1,000 minutes a month on GPT 5.4 mini with Retell's own voices.

At 1,000 minutes, that is $109, plus $2 for the number, so $111 a month. The 20 included concurrent calls cover a line this size.

An outbound reminder campaign

This is a reminder campaign: 5,000 dials a month at about 2 minutes each, so 10,000 minutes. It runs GPT 4.1 with an ElevenLabs Flash voice, and every call shows a branded caller ID.

The minutes cost $1,550. Batch calling adds $25 (5,000 x $0.005), and branded caller ID adds up to $500, counting every dial (5,000 x $0.10). A verified number ($10) and the number itself ($2) bring the month to $2,087. At that size, Retell's calculator points you to an Enterprise quote.

A clinic line under HIPAA

This is a clinic's inbound line: about 1,000 minutes a day (30,000 a month) on GPT 4.1, with a knowledge base and personal-detail removal. At peak it takes up to 30 calls at once.

The minutes cost $4,350. Ten concurrent calls above the included 20 add $80, and the number $2, for $4,432 a month. Signing Retell's BAA adds nothing. If the prompt plus knowledge base text passes 4,000 tokens on long calls, the scaling rule adds to that total.

Change the model or the call length and every total moves. Treat these as a method to copy with your own numbers.

What the same calls cost on open-source Dograh

On Dograh, the platform line changes, and you pay each model and phone provider directly at its own rates.

Dograh is an open-source, self-hostable voice agent platform. It is the orchestration layer: it runs the call flow, tools and telephony, and the speech and language models are a separate layer you choose. Self-hosted, it has no per-minute platform fee, though you still pay for the server and your model and phone accounts. Dograh Cloud, the managed version, charges a 1¢ a minute platform fee and includes 10 concurrent calls.

For the same three businesses, Dograh costs less in every row: about 2 to 4 times less for the outbound campaign and the clinic line, and about half for the small line on Dograh Cloud. The main reason is that there is no $0.055 per-minute fee; in the outbound example, your own voice account also charges only for what the agent says.

How to read the table:

• "Own keys" means you pay each AI and phone provider directly, at their own prices.
• Self-hosted includes a $48-a-month server in every row. Dograh Cloud has no server cost but adds 1¢ a minute.
• Each Dograh cost is a range: the low end is short calls, the high end is long calls where the AI re-reads more of the conversation.
• The Dograh columns leave out branded caller ID and personal-detail removal, so those two extras are not compared.

At 1,000 minutes, the server is most of the self-hosted bill, and Dograh Cloud stays cheaper up to about 4,800 minutes a month. In the outbound example, the biggest gap is the voice. Retell bills ElevenLabs Flash at $0.040 for every call minute, while your own key bills per character, about $0.016 a minute when the agent talks half the time.

Over a year at 1,000 minutes a day, Retell's voice engine alone costs $20,075 (365,000 x $0.055). Dograh Cloud's platform fees on the same minutes come to $3,650, and self-hosted Dograh has none, only servers.

Our assumptions

The Dograh columns use your own keys (BYOK) at regular list prices: Deepgram Flux at $0.0077 a minute, OpenAI's per-token prices, ElevenLabs Flash at $0.04 per 1,000 characters and Twilio's US rates for the phone. Self-hosting needs at least 8 GB of RAM and 4 vCPUs; our example server costs $48 a month.

The model gets 6 requests a minute, each with 1,500 to 4,000 tokens (about 1,100 to 3,000 words) of context, half of it cached, and a 50-token reply. The agent speaks about 400 characters a minute. We price speech-to-text on the caller's audio only; if your setup transcribes both sides, add $0.0077 a minute.

Running Dograh for these examples

The clinic example works only self-hosted. Each model and phone provider that handles patient data needs its own BAA, and a BAA delegates obligations without moving the clinic's own liability. Call records stay on your servers once you self-host. The audio stays in too only when you also host the AI models; with API keys it reaches the model provider, and the phone carrier carries the call either way.

Self-hosted Dograh allows 10 calls at once by default, a setting you can raise. How many calls one minimum-size server carries depends on your models and setup, so load-test before you count on 30.

Dograh's QA runs on the AI model you choose and is billed as normal model usage, with no separate per-minute QA fee. You can analyze only a sample of calls to keep costs down, as our guide to automated QA for voice agents explains.

Self-hosted, you control the server and choose when to upgrade. Dograh Enterprise adds support with SLAs if you need it. Pre-recorded clips for fixed lines can shrink the voice line further, as shown in How to Cut Voice AI Costs by Mixing Pre-Recorded Audio With TTS.

For every row beyond cost, read Dograh vs Retell AI. For licences, see our open-source alternative to Retell AI guide, and for the switch, how to move from Retell AI to open-source voice AI. The code is on Dograh's GitHub under BSD-2-Clause, so you can price your own setup before a single call moves.