ElevenLabs unveiled 13 new products and upgrades at its New York Summit on 11 November, pushing the voice-AI company beyond speech generation into enterprise software for customer conversations and employee work.
Key Points
- ElevenLabs announced 13 new products and upgrades at its New York Summit on 11 November 2026.
- Businesses will be able to fine-tune voice and language models on their own conversation data, with the work starting in the coming months.
- ElevenAgents gains Memory, Routines and an Architect builder, while a new ElevenAI app turns meetings into notes and next steps.
- NVIDIA’s Nemotron 3.5 Super VL joins ElevenAgents, with 15 MW of new Vera Rubin compute capacity planned.
- A Partner Program with more than 550 partners in 44 countries and a new Marketplace will carry the products into large enterprises.
What happened
The announcements rest on three voice models that arrived over the past month — Scribe v3, which transcribes speech; Eleven v4, which speaks written text aloud; and Eleven Realtime, built for live two-way voice agents. ElevenLabs describes all three as leaders on independent benchmarks.
The company is now pushing those models into specific industries. It launched Scribe v3 Medical, built for healthcare wording, which can tell apart easily confused drug names such as cycloserine and cyclosporine, and said a financial-services model is in development.
The biggest change is customisation. In the coming months, a small group of large customers will sit down with ElevenLabs researchers to tune open-source language models and the company’s own voice models on recordings and transcripts from their own live calls, then keep improving them as more conversations come in. The tuning can target whatever a business measures most — how happy callers are, or how often the agent gets it right. Speaking on stage, chief executive Mati Staniszewski said: “We believe in a future where each of you will have your own RL stack.”
ElevenLabs also previewed Interaction Engine, a system that runs a conversation rather than just generating speech. It runs the flow of a call — who speaks when, how interruptions are handled — and decides which model takes each stretch of the exchange. Routine questions go to a cheaper model; only the hard parts call on the most powerful one. “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,” Staniszewski said.
The company’s agent platform, ElevenAgents, gets three upgrades. Memory gives an agent the full history of a returning customer — what they bought, what they asked for before, the language they prefer, the hours they can talk — so nobody has to repeat themselves on the next call. Routines take on the chores around a call: screening an applicant before the conversation, booking the follow-up, sending the report, or updating the customer file once the call ends. Architect works like an in-house specialist: teams describe the agent they want in plain words, and it drafts the changes, runs them in a simulation and waits for a human to approve before anything goes live.
For employees rather than customers, ElevenLabs opened early access to ElevenAI, an app that sits in on spoken conversations — face-to-face meetings included — and writes them up as notes, summaries and action items. It can also offer live help during a call. The company says the app works with notice and consent mechanisms for everyone in the room, and early access is free.
To get the products into large organisations, ElevenLabs launched a Partner Program that opens with a network the company puts at over 550 partners across 44 countries — among them Google Cloud, IBM, Deloitte, DXC and Genesys — and an alliance with McKinsey to help companies build and scale agents. A new Marketplace lists ready-made agents from partners, aimed at sectors from healthcare and finance to insurance, education and retail, so companies can adopt one instead of building from scratch. Paying for those apps against a company’s existing ElevenLabs contract is due to follow in 2027.
Why it matters
The package is ElevenLabs’ clearest move yet from a voice toolmaker to an enterprise AI platform. Staniszewski framed the shift bluntly, saying that “within a few years, conversational agents will become critical infrastructure for every enterprise, as essential as cloud or payments.” The pitch is that every business’s conversations are different, so the models serving them should be too — a bet that customisation, not a single general model, is where the value sits.
The NVIDIA deal gives that pitch weight. NVIDIA’s open reasoning model Nemotron 3.5 Super VL now runs inside ElevenAgents, and ElevenLabs will add 15 MW of new computing power built on NVIDIA’s Blackwell Ultra hardware and its Vera Rubin platform. ElevenLabs puts the model’s accuracy at the top of its class while claiming costs 70% below rival open models on key tests — figures from the company itself, not an outside check.
The showcase customer is a government one. ElevenLabs is building an on-premises voice agent for Centralna e-Rejestracja, the booking system in Poland’s national e-Health platform, a system that books roughly 40 million appointments a year, of which 10% to 20% go unattended. It will ring patients — landline users included — with appointment reminders and offer a new time when the old one no longer works, while patient data stays on government systems. It is the kind of deployment that shows voice agents moving from call centres into public services — and Databricks made a similar pitch this week with Genie, its always-on AI coworker.
There are limits to what was announced. Customisation is still a plan for a chosen few, not a product on sale, and ElevenLabs gave no prices. Memory and Routines are also still rolling out. And the benchmarks and cost figures are the company’s own, so the real test is whether customers see the same gains in production. OpenAI launched business profiles inside ChatGPT this week, a reminder that every major AI company is now chasing the same enterprise buyers.
What happens next
The test of the summit is delivery. The first enterprise tuning projects are due to get under way in the coming months; ElevenAI’s free early access will show whether workers want an app listening in their meetings; and the Marketplace moves from browsing to buying in 2027. The Poland deployment will be the one to watch: if a national health system can run voice agents at 40 million appointments a year, other governments will follow.
Founded in 2022, ElevenLabs was reported to be worth roughly $22 billion after NVIDIA took a strategic stake in 2025. Its summit message was that voice agents are about to become infrastructure — and that the companies that shape the models to their own data will keep the advantage. “Our mission at ElevenLabs is to solve human-technology interaction, and for the first time, it’s solvable,” Staniszewski said.
Sources: Mumbrella, 12 November 2026; Unite.AI, 11 November 2026; Blockchain.News, 11 November 2026; Crypto Briefing, 11 November 2026; The Agent Times, 11 November 2026. Published 12 November 2026.



