aipopdb

19 Nov 2026  ·  9,164 works · 1,132 news indexed

News / ElevenLabs previews model fine-tuning, a workplace app and a partner program at its New York Summit

ElevenLabs previews model fine-tuning, a workplace app and a partner program at its New York Summit

AIPopDB Newsroom · via elevenlabs.io · · 4 min read

Read the original ↗
Image for ElevenLabs previews model fine-tuning, a workplace app and a partner program at its New York Summit

Image: ElevenLabs

ElevenLabs used its New York Summit on 11 November 2026 to set out a broad push into the business market: new models, more capable agents and more ways to deploy them inside a company. Chief executive Mati Staniszewski gave the keynote and described the company's mission as solving how people and technology interact, and his argument was that conversational agents are heading toward becoming core infrastructure for large organisations rather than a novelty. The company also wants to answer a worry that AI flattens everything into sameness, by helping firms keep what sets them apart. Most of what was announced comes from the company's own account, and several items are previews rather than products customers can use today. ElevenLabs began in voice and says it has built its models around natural conversation, which is the thread running through everything shown on the day.

The headline item is customisation. In the coming months ElevenLabs says it will help businesses fine-tune certain open-source language models, and its own voice models, using data from their live deployments. Customers choose the metric they want improved, such as customer satisfaction or reliability, and work alongside ElevenLabs researchers. The company likens the result to every business running its own reinforcement-learning stack. Alongside it comes Scribe v3 Medical, a healthcare transcription model built to tell look-alike drug names apart, such as cycloserine and cyclosporine. A finance-focused model is said to be in development. Scribe v3 Medical is the company's earlier example of a model made for one industry, and the new offer extends that idea to each customer's own data.

“One of the hallmarks of a great conversation with a person is that they remember what you told them and act on it. Agents should do the same”

Mati Staniszewski, CEO and co-founder of ElevenLabs, at the ElevenLabs New York Summit · elevenlabs.io ↗

On the agent side, a previewed Interaction Engine is meant to handle turn-taking and interruptions in a conversation and to pick a different language model for each stage: a cheaper one for triage, a frontier one for the hard steps. Two features for ElevenAgents are also forthcoming. Memory lets an agent recognise an authenticated returning customer and recall earlier exchanges, with the customer's CRM remaining the primary record, so the agent supplements the company's files and does not replace them. Routines chain actions before, during and after a call, for example pre-screening a mortgage applicant and phoning back within minutes. Yahoo, Westgate and KPN were named as companies already using ElevenAgents.

For staff rather than customers, Staniszewski previewed ElevenAI, an app that listens to spoken conversations, including meetings held in person, and turns them into notes and action items. It also does dictation. It runs on Scribe v3, described as covering 102 languages, with Scribe v3 Live as the low-latency variant, and notes are taken with participants' consent. Free early access opened the same day. The pitch is that the same speech technology behind the company's customer agents can serve everyday office work, in the room as well as on the phone, and the consent step is meant to keep recording of colleagues above board.

The go-to-market side got as much attention. A new Partner Program is aimed at consultancies, integrators, software vendors and agencies, and an alliance was announced with McKinsey and QuantumBlack, AI by McKinsey. ElevenLabs says its network already counts more than 550 partners in 44 countries, among them Google Cloud, IBM, Deloitte, DXC and Genesys. A Marketplace is launching with industry apps for healthcare, financial services, insurance, education and retail, although buying through a customer's existing ElevenLabs commitments only begins in 2027.

Infrastructure and a public-sector example rounded out the day. NVIDIA's Nemotron 3.5 Super VL is now available inside ElevenAgents, and ElevenLabs is adding 15 MW of NVIDIA Vera Rubin compute on top of its Blackwell Ultra systems. In Poland, an on-premises voice agent is planned for the national e-Health appointment system, Centralna e-Rejestracja, with EasyCall supporting. It will phone patients who only have landlines. According to ElevenLabs, the system handles about 40 million appointments a year, with 10 to 20 percent of them going unattended, which is the gap the calls to landline-only patients are meant to narrow.

What happens next is largely a matter of delivery. Fine-tuning, the Interaction Engine, Memory and Routines are all due in the coming months rather than shipped, and the Marketplace purchasing route is a 2027 item. An analysis on Unite.AI, itself written by an AI agent and largely a rewrite of the company's post, concludes that the strategy now depends on moving previewed products into production and turning Marketplace browsing into purchases. For the voice-AI field, the summit shows ElevenLabs moving from the voices it began with toward selling whole conversational systems to enterprises.