ElevenLabs Realtime: Two New Voice-Agent Models, What They Cost and What They Mean for Small Businesses
ElevenLabs Realtime launched on 20 Oct 2026: two speech-to-speech models, Mini and Max, for AI voice agents. What each one is for, the $0.08-a-minute list price plus AI model usage, who can use it yet, and what it means for an AI phone agent.
Research-based overview: we have not tested this hands-on yet, so there is no score. Facts come from the makers' own pages, linked below.
On this page
Key takeaway: ElevenLabs Realtime is a new pair of voice-agent models, launched on 20 October 2026, that listen and talk back in one step. Both are listed at $0.08 a minute plus AI model (LLM) usage, in US dollars. They make an AI phone or web voice agent simpler to build. They are not a ready-made receptionist, and they are rolling out gradually, so your account may not have them yet. We read ElevenLabs’ own pages on 22 October 2026 and haven’t made a test call yet.
- Two models: Eleven Realtime v1 Mini for everyday conversations at low cost, and v1 Max for complex, multi-step tasks with “heavy tool calling” (looking things up or taking actions during a call). (ElevenLabs blog)
- Price: $0.08 per minute plus LLM usage for each model. ElevenLabs says Max “costs more than Mini”, because its token rates are higher. (API pricing, docs)
- Languages and voices: more than 90 languages and any of the 17,000+ voices in the ElevenLabs library. (ElevenLabs blog)
- Quality: ElevenLabs publishes benchmark results for both models. They’re the company’s own reported numbers, shown below, not ours.
Research-based news. We haven’t tested Eleven Realtime yet. Every fact here comes from ElevenLabs’ own pages, checked 22 October 2026, with prices rechecked 23 October 2026. All prices are in US dollars (USD) and exclude taxes.
What happened: ElevenLabs Realtime in plain words
On 20 October 2026, ElevenLabs published “Introducing Eleven Realtime: a simpler way to build natural voice agents”. It calls Eleven Realtime “a new class of Speech to Speech models” and says “each model is a complete voice agent”. (ElevenLabs blog)
Speech to speech means the model hears the caller and answers out loud itself. Until now, a voice agent was usually a chain of separate parts: speech recognition (turning speech into text), a language model (writing the reply), and text to speech (reading it out). You had to choose and tune each part. ElevenLabs’ pitch is that Realtime handles that chain for you.
Mini vs Max in one table
| Eleven Realtime v1 Mini | Eleven Realtime v1 Max | |
|---|---|---|
| What ElevenLabs says it’s for | “Natural conversations at low cost”: AI assistants, teachers and interactive characters | Complex tasks: customer support, lead qualification and multi-step workflows |
| How it answers | “Lowest latency and cost”; the default Realtime model | “Reasons before it responds” |
| Tool use (lookups and actions mid-call) | Supported | Recommended “for use cases with heavy tool calling” |
| Base price | $0.08/min + LLM usage | $0.08/min + LLM usage |
| Model ID (for developers) | eleven_realtime_v1_mini |
eleven_realtime_v1_max |
Sources: ElevenLabs blog, Realtime models docs, API pricing, prices rechecked 23 Oct 2026.
What it costs
ElevenLabs’ API pricing page (checked 23 Oct 2026) lists both models as “New”, each at $0.08 + LLM usage, price per minute. The same page shows the token rates that make up the “LLM usage” part:
| Model | Connected audio, per extra minute | Input tokens, per 1M | Output tokens, per 1M |
|---|---|---|---|
| Realtime v1 Mini | $0.080 | $0.20 | $6.00 |
| Realtime v1 Max | $0.080 | $1.80 | $9.00 |
Tokens are the small chunks of text an AI model reads and writes. A longer call, or one with more lookups, uses more of them. ElevenLabs also lists burst pricing of $0.16 a minute for audio above your plan’s limit on calls running at the same time. (API pricing)
Plans (monthly, USD, from the ElevenLabs pricing page, checked 23 Oct 2026): Free $0 (10k credits a month), Starter $6 ($1 for the first month until 1 November), Creator $22 ($11 for the first month), Pro $99, Scale $299 and Business $990. The API pricing page shows estimated included Realtime minutes for each plan, starting at about 14 minutes of Mini or 12 of Max in its “Free / Pay as you go” column, and notes that “actual minutes vary with conversation length and tool use”. Whether a new free workspace can use Realtime today is unconfirmed, because of the gradual rollout.
Example calculation, not a quote: a 3-minute call at the base rate is 3 × $0.08 = $0.24, before LLM token costs and before any phone-number or telephony charges. Telephony costs are unconfirmed; we haven’t checked them.
How good is it? ElevenLabs’ reported results
These are ElevenLabs’ own reported numbers, not our tests. ElevenLabs says Artificial Analysis evaluated both models independently on public benchmarks. We haven’t checked these results ourselves. (ElevenLabs blog)
| Benchmark (as described by ElevenLabs) | v1 Max | v1 Mini |
|---|---|---|
| Big Bench Audio: spoken reasoning, 1,000 questions | 98.8% correct | 75.9% correct |
| Average time until it starts speaking | 0.57 s | 0.62 s |
| Full-Duplex-Bench: pauses, turn-taking, interruptions | 94.1% | 95.8% |
| Speech Agent Arena: blind human preference score | 998 | 988 |
| Tool-use tasks completed | 87.1% | 60.3% |
By ElevenLabs’ own figures, Max is ahead on reasoning and tool use, and Mini is slightly ahead on conversational flow. Neither number tells you how it will handle your customers, your accent mix or your booking system.
Who can use it now
- Gradual rollout: “Realtime models are rolling out gradually.” If the Voice mode setting doesn’t appear on an agent’s Agent tab, the models aren’t available for your workspace yet. (docs)
- Inside ElevenAgents (ElevenLabs’ agent builder): each voice agent has a Voice mode, Custom or Realtime. In Realtime mode, ElevenLabs manages the language model, voice model and speech recognition for you. The system prompt, voice, tools, knowledge base and turn settings work the same in both modes. ElevenLabs recommends this route; it includes phone, web, WhatsApp and contact-centre channels. (docs, blog)
- For developers: the standalone Realtime API follows OpenAI’s Realtime API specs, so code built for OpenAI’s Realtime SDK can point at ElevenLabs instead. It “connects over WebSocket only”, so a WebRTC setup has to move to WebSocket first. It runs separately from ElevenAgents. (blog, compatibility docs)
- Data location and compliance: Mini is available in ElevenLabs’ US, EU, India and Singapore data-residency environments. Max is available in the US and EU but not in India or Singapore. For Zero Retention Mode and HIPAA (a US health-privacy law), only Max is available, in the US and EU. (docs)
What it means for a small business
For a clinic, salon, studio or consultancy that misses calls, this makes the building blocks of an AI phone agent simpler and cheaper to try. It could answer common questions, take booking requests or qualify leads, in 90+ languages.
It is still a building block, not a finished front desk. You would still need:
- A script covering what the agent says, what it must never promise, and when it hands over.
- Tools connected to your calendar or booking system so it can actually book, plus a phone number.
- Testing in your customers’ languages, including interruptions and background noise.
- A human hand-off for complaints, emergencies and anything unclear.
If you want to compare ready-made options, see our AI receptionist comparison. Prefer a front desk that’s set up for you? AIFix Automation sets up AI assistants for bookings, reminders and reviews on accounts in your name. That’s our own service.
What we checked, and what we’ll test
We read the launch post, the API pricing page, the plans page and the Realtime docs on 22 October 2026. We have not made a test call yet. When we do, we plan to:
- Check whether Voice mode appears on a new free ElevenAgents workspace.
- Have a 2-minute booking conversation with Mini and with Max, in English and in one other language, with an interruption in the middle.
- Note the cost the dashboard shows after each call.
We’ll add an update line here if the rollout or prices change.
FAQ
What is ElevenLabs Realtime?
A family of two speech-to-speech models, Eleven Realtime v1 Mini and v1 Max, that ElevenLabs launched on 20 October 2026 for building AI voice agents. (ElevenLabs blog)
How much does it cost?
ElevenLabs lists both at $0.08 per minute plus LLM usage, in USD, and says Max costs more than Mini overall because of its higher token rates. (API pricing, checked 23 Oct 2026)
Can I use it on the free plan?
Unconfirmed. ElevenLabs has a Free plan, and its API pricing table shows estimated Realtime minutes for it, but the models are rolling out gradually. Check for the Voice mode setting in your own workspace.
Is it a ready-made AI receptionist?
No. It’s the voice model underneath. You still need a script, booking tools, a phone number and a human backup.
Have you tested it?
Not yet. This post is based on ElevenLabs’ published pages, and the benchmark numbers are ElevenLabs’ own.
Sources
- Introducing Eleven Realtime: a simpler way to build natural voice agents (20 Oct 2026), ElevenLabs, 2026 (checked 22 Oct 2026)
- API pricing, ElevenLabs (checked 23 Oct 2026)
- Realtime models (docs), ElevenLabs (checked 22 Oct 2026)
- Pricing (plans), ElevenLabs (checked 23 Oct 2026)
- OpenAI Realtime API compatibility (docs), ElevenLabs (checked 22 Oct 2026)
- AIFix Automation, AIFix (checked 23 Oct 2026)
This post was drafted with help from an AI assistant, then checked, edited and approved by Numan Malik.
Numan Malik, Founder, AIFix. Reviewed and approved by Numan before publishing. More about Numan · Editorial policy
