LATEST
AI Tools

Retell Tell-1: A $0.035-a-Minute Model Built for Live Calls

Updated November 17, 2026
12 min read
Editor's Note: This article was human-reviewed for accuracy
A person holding a call-center headset with a microphone

Quick Answer & Key Takeaways

  • • Retell launched Tell-1 on November 12, 2026, a language model built for live contact-center calls at $0.035 a minute.
  • • Retell says it responds in under 100 milliseconds on its own benchmark and beats frontier models on live-call tasks.
  • • The headline rate is only the model line, though — the full per-minute cost is higher once voice, infrastructure, and telephony are added.

Most voice agents fail in the same place: the messy middle of a real phone call. A caller gives half an answer, changes the subject, or asks the agent to look something up while the line goes quiet, and the model either guesses, restarts, or stalls. Retell's answer to that problem is Tell-1, a language model launched on November 12, 2026, built specifically for live contact-center calls. It costs $0.035 a minute, responds in under 100 milliseconds on Retell's own benchmark, and, according to Retell, beats the latest models from OpenAI, Google, and Anthropic on the five things that matter most on a live call. That is a strong claim from a company that also sells the platform the model runs on, so the benchmark deserves a hard look before the price does. This review breaks down what Tell-1 actually is, what a minute of it really costs once the rest of the stack is added, and where the honest limits sit.

What Tell-1 is, and what it is not

Tell-1 is a language model, not a voice agent on its own. It is the thinking layer inside a Retell AI voice agent: the part that decides what to say and when to act, while separate speech-to-text and text-to-speech components handle listening and speaking. You pick it per agent from Retell's model menu, alongside models from OpenAI, Anthropic, and Google, which means it is a choice inside Retell's platform rather than a standalone product you can take elsewhere.

The design idea is what Retell calls interleaved reasoning. Instead of treating each turn as an isolated question, the model reasons through the conversation as it unfolds: which of the business's rules apply now, which do not, what information is still missing, and whether to ask, respond, or act. In practice, Retell says this helps agents follow long, complex instructions from the first minute of a call to the last; work with partial answers by asking for the missing detail instead of guessing; handle detours by acknowledging an off-topic question and returning to the task; and call the right tool, such as an availability or account lookup, once it has enough information.

The training data behind that pitch is large and, so far, unverifiable from outside. Retell says Tell-1 was trained on hundreds of millions of calls, and launch coverage reports more than one million hours of customer calls behind it. Both figures are the company's own, with no independent audit published. What is verifiable is the company's scale: Retell, founded in 2023, says it serves more than 4,000 businesses, including Lenovo, CVS Health, and BILL, handles more than 100 million connected calls a month, and runs at a $100 million annualized revenue run rate.

Two limits belong in the open before the price does. First, Tell-1 is in beta: Retell's pricing page lists it as "Tell 1 Beta," and no launch date for general availability has been published. Second, Retell publishes no model size, no weights, and no way to run Tell-1 outside its platform. If you want a model you can self-host or plug into another stack, this is not it. For a closer look at a voice model you can run through an API instead, see our Mistral Voxtral 3 review.

The price, line by line

The headline number is real, but it is only one line of the bill. Retell's pricing page lists Tell-1 Beta at $0.035 per minute for voice calls and $0.011 per AI message for chat agents. That is the language-model line. A live call needs three more paid components from the same page: Retell Voice Infra at $0.055 a minute, a voice at $0.015 a minute for Retell's own platform voices, and telephony at $0.015 a minute. Add them up and a Tell-1 minute costs about $0.12 before any add-ons, or about $7.20 an hour. That sits inside Retell's own quoted range for a full voice agent, $0.07 to $0.31 a minute, and it is the number to budget against, not $0.035.

At November 17, 2026 market rates (1 USD = 0.758 GBP, 1.429 CAD, approximate), the model line works out to about £0.027 a minute or C$0.050 a minute, roughly £1.59 or C$3.00 an hour. The full stacked minute is about £0.091 or C$0.171, roughly £5.46 or C$10.29 an hour. Retell gives new accounts $10 in free credits, about £7.58 or C$14.29, with full platform access and no contract.

Against the other models on Retell's own pricing page, Tell-1 sits in the middle, not at the bottom. The table below uses Retell's published per-minute LLM rates:

Model (on Retell)Per minuteWhat it is
GPT 6 Luna$0.0032OpenAI's smallest model on the menu
Claude 5.5 Haiku$0.0032Anthropic's smallest model on the menu
GPT 6.1 Terra$0.032OpenAI's mid-tier model, Retell's recommended pick
Tell-1 Beta$0.035Retell's own contact-center model
GPT 6 Sol$0.064OpenAI's larger model
Claude 5.5 Sonnet$0.064Anthropic's mid-tier model
Gemini 4 Argon$0.128Google's flagship on the menu

Two readings follow from that table. Tell-1 is about 45% cheaper than GPT 6 Sol or Claude 5.5 Sonnet, and about 73% cheaper than Gemini 4 Argon, which is a genuine saving if it holds up on live calls. But it is not the cheapest model on Retell's own menu: GPT 6.1 Terra undercuts it by $0.003 a minute, and the small Luna and Haiku models cost less than a tenth of it. The honest framing is that Tell-1 is priced as a mid-tier specialist, not as a budget option, and any "cheapest" claim about it would be wrong on Retell's own price list.

The add-ons and billing rules are where a monthly invoice can drift from the headline. Knowledge Base adds $0.005 a minute, AI Quality Assurance costs $0.10 a minute after the first 100 free minutes, PII removal adds $0.01 a minute, and Safety Guardrails add $0.005 a minute. A Retell phone number is $2 a month, extra concurrency beyond the 20 included calls is $8 per concurrent call a month, and SMS is $0.01 a message. Calls are tracked to the nearest second, but two rules inflate short or prompt-heavy calls: a call under 10 seconds with a dynamic greeting is billed as 10 seconds, and a prompt averaging over 4,000 tokens inflates billed time in proportion, so a 4,800-token prompt makes a 60-second call bill as 72 seconds. Billing covers the full duration of a call, including silence, because the speech-to-text engine stays active throughout. On the fair side, only connected calls are billed, a failed connection costs nothing, and once a call transfers to a human the AI fee stops and only telephony continues. There is no platform fee, no monthly minimum, and no annual contract; enterprise pricing is custom.

A worked example makes the stack concrete. A team running 10,000 Tell-1 minutes a month with Retell platform voices pays $350 for the model, $550 for voice infrastructure, $150 for the voice, and $150 for telephony: $1,200 a month before add-ons, about £910 or C$1,715. The same 10,000 minutes on Claude 5.5 Sonnet would cost $640 for the model line alone, so Tell-1 saves about $290 a month on the LLM line at this volume, before any difference in call outcomes. Support teams comparing per-minute pricing with per-resolution pricing should also read our Zendesk Outcome AI Platform review, which prices resolved issues rather than minutes.

What the benchmark actually proves

Retell's headline benchmark claim is that Tell-1 posted the highest overall score of any model tested across the five measures that matter most on a live contact-center call: instruction following, grounding, tool calling, colloquial quality, and latency, ahead of the latest models from OpenAI, Google, and Anthropic. The benchmark is called vCX, and it lives on Retell's own site at bench.retellai.com. That provenance is the catch. Retell built both the model and the test, and no independent verification of the vCX results exists yet; launch coverage notes the same gap. Retell also publishes no per-model scores or methodology document alongside the ranking, so a buyer cannot check how close the race was or which measure decided it.

There is one independent data point, but it measures the platform, not Tell-1. In an independent Cekura benchmark of six platforms, Retell AI scored the highest workflow accuracy at 95.7%, passing 22 of 23 evaluators in every repetition. That is a real third-party result for Retell's platform, and it is worth more than a vendor-run leaderboard. It is not, however, evidence about Tell-1 itself, and the two should not be conflated.

Latency needs the same care. Retell says Tell-1 responds in under 100 milliseconds on its live benchmark and is built to stay under 300 milliseconds even at peak call volumes. Retell's homepage separately quotes about 600 milliseconds for a voice agent's response time. Those figures measure different parts of the pipeline, and Retell publishes no single end-to-end number that reconciles them, so treat the sub-100-millisecond figure as the model's benchmark response, not as the time from a caller going quiet to a spoken reply. The gap matters because latency is the whole pitch: Retell's CEO, Bing Wu, puts it plainly in the launch release, saying the bottleneck for voice AI agents "isn't intelligence but timing: the model has to reason through the business's rules and act in the time it takes a person to pause."

Retell Dialer, launched the same day

Tell-1 did not launch alone. Retell also introduced Retell Dialer, described as the first dialer for voice agents to orchestrate outbound campaigns across AI voice and SMS agents. Where Tell-1 handles decisions during a live conversation, Dialer turns the outcome of that conversation into the next action: a requested callback replaces the next planned attempt and resumes the same conversation at the contact's requested local time, a qualified prospect moves to a specialist agent, and someone interested but not yet ready receives an SMS follow-up from a nurture agent. The feature list covers contact screening, local presence and calling windows, number rotation and spam monitoring, configurable retries, and CRM integration.

Dialer is also where the compliance fine print bites. Outbound calling in the United States requires prior consent under the TCPA, some states require disclosure that the caller is automated, and call recording consent rules vary by state. Retell's own FAQ flags California's AB 1609, which takes effect January 1, 2027 and sets requirements for how large businesses connect customers to a human. None of this is a reason to avoid outbound AI calling, but it is a reason to check consent and disclosure rules with counsel before a first campaign. This is general information, not legal advice.

Who should use Tell-1, and who should wait

Tell-1 fits contact centers already running on Retell, or evaluating a voice-agent platform, where per-minute LLM cost is a real line item and calls are long enough for reasoning to matter. High-volume inbound support, appointment scheduling, and collections are the natural use cases: the model is built for exactly the partial answers, detours, and mid-call lookups those calls produce, and the $10 in free credits plus simulation testing lets a team try it on its own scripts before committing.

Wait if you need a model you can run outside Retell, since Tell-1 is platform-locked and in beta, with no published timeline for general availability. Wait as well if your calls are very short, because the 10-second minimum and prompt-size billing inflate short calls, and if you need independent benchmark proof before switching models, since none exists yet for Tell-1 itself. Teams whose main need is accurate transcription rather than live conversation should compare the speech layer separately; our ElevenLabs Scribe v3 review covers a transcription model at $0.20 an hour.

Common mistakes

The most common mistake is treating $0.035 a minute as the total cost of a Tell-1 call. It is the language-model line only; the full minute runs about $0.12 with Retell's voice infrastructure, a platform voice, and telephony. The second is comparing Tell-1's LLM line with a competitor's all-in per-minute price, which flatters Tell-1 by leaving out its own stack. The third is ignoring prompt size: a prompt averaging over 4,000 tokens inflates billed time, so a bloated system prompt quietly raises the bill on every call. The fourth is reading the vCX number-one ranking as independent proof when it is Retell's own benchmark. And the fifth is launching outbound campaigns on Dialer without confirming consent and disclosure rules for each jurisdiction called.

How much does Tell-1 cost?

Tell-1 Beta costs $0.035 per minute for voice calls and $0.011 per AI message for chat agents. A full voice minute with Retell's voice infrastructure, a platform voice, and telephony costs about $0.12 before add-ons. New accounts get $10 in free credits.

Is Tell-1 available now?

Yes, in beta. You select Tell-1 as the language model for an agent inside Retell's platform. It is not available as a standalone API or for use outside Retell, and Retell has not published a general-availability date.

Is Tell-1 faster than GPT or Claude on live calls?

On Retell's own vCX benchmark, Tell-1 ranked first overall and responds in under 100 milliseconds, ahead of the latest models from OpenAI, Google, and Anthropic. No independent benchmark of Tell-1 exists yet, so that ranking is Retell's claim, not third-party proof.

Can I use Tell-1 with my own phone number?

Yes. Retell connects to existing telephony through SIP trunking, so you keep your number and carrier, or you can buy a Retell number for $2 a month.

What does Retell Dialer do?

Retell Dialer runs outbound campaigns across AI voice and SMS agents. It handles callbacks at the contact's requested time, routes qualified prospects to specialist agents, sends SMS follow-ups to contacts who are interested but not ready, and includes contact screening, local presence, calling windows, number rotation, and spam monitoring.

Does Tell-1 work for chat as well as voice?

Yes. Tell-1 is also listed for chat agents at $0.011 per AI message. The launch focuses on voice, where the latency and interleaved-reasoning design matter most, but the same model line is priced for chat on Retell's pricing page.

The bottom line

Tell-1 is a credible specialist bet rather than a bargain. Its $0.035-a-minute model line undercuts the flagship general models on Retell's own menu by a wide margin, its training is specific to the messy reality of live calls, and the company behind it handles more than 100 million calls a month. The honest budget is about $0.12 a minute for the full stack, not $0.035, and the benchmark claim is something to test on your own traffic, which Retell's $10 in free credits and simulation testing are built for. If the sub-100-millisecond responses hold up on your scripts, the saving is real. If they do not, the cheaper model on the menu is GPT 6.1 Terra, one line above it.

Sources

12 min read
101 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more