ElevenLabs announced its Interaction Engine at its New York Summit on November 11, 2026, as the centrepiece of 13 new products and upgrades. It is a system that manages a whole voice conversation — deciding when to speak, when to listen, and which AI model should handle each moment — instead of running every reply through one expensive model. Most of it is still in preview, and it is aimed at businesses first, but it changes what ElevenLabs' free voices will sound like next.
Quick Answer: The Interaction Engine in One Box
The Interaction Engine is ElevenLabs' new orchestration layer for voice agents. It handles turn-taking and interruptions, sends routine questions to cheaper models, and hands complex moments to stronger models — so a customer-service voice agent sounds natural without burning money on every word. It launched in preview at the November 11, 2026 Summit alongside three new models: Scribe v3 for transcription, Eleven v4 for text-to-speech, and Eleven Realtime for speech-to-speech agents. If you just want to hear text read aloud free, nothing changes today — but the voices behind the free tier are about to get noticeably better.
What actually happened at the Summit
ElevenLabs, the AI voice company, announced 13 new products and upgrades at its New York Summit on Wednesday, November 11, 2026. The package falls into four areas: custom models tuned on a company's own live data, voice agents that remember returning customers and automate follow-up work, a new workplace app for meetings, and a partner channel backed by Nvidia infrastructure.
The honest context first: most of the new items are previews, early access, or phased rollouts rather than finished products you can use today. So the immediate effect is a roadmap, not a new catalogue — and the claims below are ElevenLabs' own, not independently tested yet.
What the Interaction Engine actually does
A real conversation is not one job — it is dozens of small ones. Saying "hello" and checking an order status are easy. Handling a billing dispute or a refund is hard. Running every moment through the strongest model is like hiring a senior lawyer to answer the phone: it works, but the bill is absurd, because every second of conversation burns tokens twice — once to hear, once to think.
The Interaction Engine splits the work. A cheap model handles triage and routine answers. A frontier model takes the complex parts. The engine manages turn-taking and interruptions in between, so the caller is not talked over or left hanging.
As ElevenLabs co-founder and CEO Mati Staniszewski put it at the Summit: "Interaction Engine orchestrates all this under the hood, so you can be confident you're delivering the right outcome for the customer, without wasting money on tokens."
The part nobody has proven yet: ElevenLabs has not published an independent benchmark showing how much businesses actually save, or whether the system resolves more customer problems than the setup it replaces. Treat the savings story as a promise until real deployments report back.
The three new models behind it
Three voice models shipped around the Summit, and each one does a different job:
| Model | What it does | The number that matters |
|---|---|---|
| Scribe v3 | Speech-to-text transcription | 102 languages; ElevenLabs says 27% fewer word errors than Scribe v2 on its own evaluation |
| Eleven v4 | Text-to-speech | Reads the tone, pacing, and context of a script for more expressive speech |
| Eleven Realtime | Speech-to-speech | A new class of models built for fast, accurate voice agents |
Scribe v3 also comes in a Medical version tuned for healthcare terms, and ElevenLabs says industry-specific versions for areas like financial services are coming. The 27% error reduction is ElevenLabs' own measurement, not an independent test — worth knowing before you repeat the number.
Eleven v4 is the one most readers will actually hear. It is the text-to-speech model that powers the voices, and its job is to read a script the way a person would — speeding up, slowing down, and shifting tone with the words. Our ElevenLabs v4 free guide covers how to try the current voices without paying.
Agents that remember you — and a workplace app
Alongside the engine, ElevenLabs upgraded ElevenAgents, its platform for voice and chat agents. Two changes matter:
- Memory gives agents the full context of every returning customer — what they said before, from a new shirt size to a pending house move — so nobody has to repeat themselves.
- Routines add multi-step automation, so an agent can run a whole follow-up sequence (a callback, a text, a CRM update) instead of handling one call and stopping. Fine-tuning models on a company's own live data is also coming for enterprise customers — ElevenLabs' own timeline puts it in the months ahead, for selected enterprises first.
The company also opened early access for ElevenAI, a new app that acts as an AI teammate for enterprise teams, starting with meetings: cross-app voice dictation, meeting notes with per-attendee action items, real-time call guidance, and handoffs between human staff and automated agents.
On the infrastructure side, ElevenLabs announced a Partner Program with more than 550 partners in 44 countries, an alliance with McKinsey to help organisations build and scale agents, and an expanded Nvidia partnership: Nemotron 3.5 Super VL is now available inside ElevenAgents, and ElevenLabs will add 15 MW of new Vera Rubin compute capacity. ElevenLabs says Nemotron delivers leading customer-support task completion at 70% lower cost than rival open models — again, the company's own claim, not an independent result.
What it means for you if you just want free voices
Here is the straight answer: almost nothing changes today, and that is fine.
- If you use ElevenLabs' free tier, the voices will gradually get better as v4 rolls out, but the free tier stays capped — the well-known 10,000-character limit is still the wall. Our ElevenLabs alternatives roundup compares seven free tools with bigger limits.
- If you build voice agents, the Interaction Engine is the news that matters — but it is a preview for enterprise teams, not a button you can press today.
- If you just want text read aloud, you do not need any of this. A free browser tool turns your text into speech in seconds, with no account and no character meter to watch.
The bigger picture is the bet ElevenLabs is making: the winner in voice will not be the company with the single best model, but the one that orchestrates many models well and remembers who it is talking to. That is a reasonable bet — and the Interaction Engine is the clearest version of it yet.
How to hear the new voices free today
You cannot try the Interaction Engine itself — it is a behind-the-scenes system for businesses. But you can hear where the voices are heading right now:
- Try ElevenLabs' free tier first. Sign up, paste a paragraph, and pick a v4 voice. Listen for the pacing and tone shifts — that is the new model at work. Watch the character counter; the free allowance runs out fast.
- If you hit the limit, use a free reader. TextToSpeechH turns up to 10,000 words into natural speech in 30 voices and downloads the MP3 free, no sign-up. It is the fastest way to hear a whole document read aloud while ElevenLabs' new models are still rolling out.
- If you want to build with voice, our ElevenLabs Realtime voice agent guide explains how the agent side works, and our AI voice cloning guide covers cloning — including the legal rules in our is AI voice cloning legal guide.
One practical note: do not switch tools or pay for a plan on the strength of a launch announcement. Try the voices on your own real text, measure what they actually save you, and judge the claims against your own results before you pay or switch.
The bottom line
The Interaction Engine is ElevenLabs' move from selling voice models to selling the whole conversation: the right model for each moment, memory of who is calling, and automation after the call ends. The models behind it — Scribe v3, Eleven v4, Eleven Realtime — are real and shipping. The savings and quality claims are not independently proven yet, and most of the package is still in preview. If you are a business building voice agents, this is the announcement to watch. If you just want free, natural voices today, the best move is the same as ever: try the free tier, and when the meter runs out, use a free reader that does not count characters.
Try it now: paste any text into our free text-to-speech tool and hear it in 30 voices, or read our best free text-to-speech tools roundup to compare the free options side by side.
By Mauhnish — I build TextToSpeechH and write these guides from what actually works when you test it, not from feature lists. If a step above didn't work the way I described, the tool changed — tell me and I'll fix the guide.
Sources & References
- W3C Web Accessibility Initiative — "Audio Content & Video Content" text-to-speech guidance: www.w3.org/WAI/media/av/av-content/
- Wikipedia — Speech Synthesis & Speech Recognition: en.wikipedia.org/wiki/Speech_synthesis
- van den Oord et al. — WaveNet: A Generative Model for Raw Audio: arxiv.org/abs/1609.03499
- Shen et al. — Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions (Tacotron 2): arxiv.org/abs/1712.05884
- Google — Neural Text-to-Speech & AI Overview documentation: cloud.google.com/text-to-speech