Introduction

ElevenLabs held its Summit NYC at Lincoln Center on November 11, 2026, and posted six blog entries during the keynote. Here's the short version for makers: there's no new public voice or music model today. ElevenLabs points back to the three models it shipped over the last month: Scribe v3 for transcription, Eleven v4 for expressive speech and Eleven Realtime for speed. What's new is the layer on top of them. There's a new desktop app, ElevenAI, with free early access opening today. ElevenAgents is getting a conversation orchestrator called Interaction Engine, plus Memory and Routines. ElevenLabs says fine-tuning of its voice models is coming in the next few months. And there's a batch of NVIDIA and enterprise-partner news. Everything here is closed and hosted by ElevenLabs.

Sources: the official recap, Everything announced at ElevenLabs Summit NYC, the ElevenAI page, the NVIDIA partnership post, the Marketplace and Partner Program posts, the enterprise customer roundup, and the

.

What You Can Use Today

ElevenAI (free early access)

  • A new desktop app that ElevenLabs calls "an AI teammate" for work. It does dictation, meeting notes, briefs and handoffs.
  • Transcription runs on Scribe v3. In meetings it takes notes "with everyone's consent" and sends each attendee a summary and action items when the call ends. It can also help live during a call.
  • It's built on ElevenAgents and draws on context from past conversations. The product demo shows a meeting assistant answering "what have we discussed so far" across six earlier calls.
  • ElevenLabs says notice and consent mechanisms are built in. Sign-up is on the ElevenAI page. There's no pricing and no platform list yet beyond "desktop app".

NVIDIA Nemotron 3.5 Super VL in ElevenAgents

  • It's now selectable as the LLM behind an ElevenAgents voice agent. On ElevenLabs' own customer-support benchmark, the company says it had leading end-to-end task completion among open models at 70% lower cost.
  • You can turn reasoning off for simple turns. ElevenLabs says that puts it among the three fastest models on the platform.

What Was Previewed (Not Shipping Yet)

Interaction Engine

  • A new orchestration system for agents that starts from "the rhythm of the conversation". It handles turn-taking and interruptions, so the agent knows when to jump in and when to stop talking.
  • It also routes each part of a conversation to a suitable LLM. A cheaper model handles triage and a frontier model steps in for harder tasks.
  • No release date, docs or pricing yet.

Memory and Routines for ElevenAgents

  • Memory: an agent keeps a structured record of each returning, authenticated customer, holding the details that never had a CRM field. ElevenLabs' examples are "prefers Spanish, only after 5 pm" and "moving in March".
  • Routines: multi-step automations that run before, during and after a call. Examples include calling a mortgage applicant back within minutes, offering a cancelled slot from a waitlist, writing to a CRM or emailing a report.

Fine-tuning for voice models and open LLMs

  • "In the coming months", ElevenLabs will help businesses fine-tune ElevenLabs voice models and certain open-source language models on data from their own live deployments. Teams pick the metrics, such as CSAT or reliability. CEO Mati Staniszewski compared it to each company having its own reinforcement-learning stack.
  • An early example: ElevenLabs and Dutch telecom KPN fine-tuned speech recognition on Dutch data, which raised identity-verification accuracy from about 50% to 90%. ElevenLabs says every Dutch deployment on the platform now gets that improvement.

Domain-specific speech-to-text

  • The recap names Scribe v3 Medical, a healthcare model it says can tell drug names like cycloserine and cyclosporine apart. A finance-tuned model is "in the works".
  • As of 17:10 UTC on Nov 11, the docs changelog only lists Scribe v2 Medical (scribe_v2_medical). There's no API entry for Scribe v3 Medical yet, so treat it as announced, not shipped.

Ecosystem and Enterprise News

  • Partner Program: ElevenLabs has more than 550 partners in 44 countries, including Google Cloud, IBM, Deloitte, DXC and Genesys. The program formalises how integrators deploy ElevenAgents. McKinsey and QuantumBlack signed a separate alliance.
  • ElevenLabs Marketplace: a curated catalogue of industry agents built on ElevenLabs. Launch partners include Salesforce, IBM, NiCE, Twilio and Zendesk, across healthcare, finance, insurance, telecoms and retail. ElevenLabs says customers will be able to buy them with their existing ElevenLabs commitment "starting next year".
  • NVIDIA compute and sovereign voice: ElevenLabs is adding 15 MW of NVIDIA Vera Rubin capacity on top of its Blackwell Ultra systems. It's also running an on-premises voice agent for appointments on Poland's national e-Health platform, Centralna e-Rejestracja, on NVIDIA Confidential Computing. The agent calls patients who only have landlines. The two companies are also exploring on-device voice for robots and vehicles.
  • Scale claims: ElevenLabs says enterprises now bring in more than 55% of its revenue, and employees at nearly 70% of Fortune 500 companies use the platform. It lists 16,000 voices and languages that 5.5 billion people speak natively.

Why It Matters

The Summit confirms where ElevenLabs is heading. The model race (Scribe v3, v4, Realtime) shipped in the weeks before the event. The keynote was about owning the interaction layer: turn-taking, memory, follow-up work and per-company fine-tuning. For voice builders, the two things to watch are Interaction Engine and voice-model fine-tuning. Interaction Engine goes straight at the turn-taking problem that open full-duplex stacks like Sophon and MOSS-Live are also trying to solve. Fine-tuning would let a brand train its own ElevenLabs voice on real calls. For solo creators, the only thing you can touch today is the ElevenAI early-access sign-up.

Caveats: Interaction Engine, Memory, Routines and fine-tuning have no dates, docs or prices. The Nemotron cost figure comes from ElevenLabs' internal benchmark. Some attendee posts mention other launches, such as a "Duplex" speech-to-speech mode. We haven't found a blog or docs entry for those, so they're not in this piece. We'll update when ElevenLabs posts them.

Conclusion

ElevenLabs Summit NYC was a platform day, not a model day. Sign up for ElevenAI if you want Scribe v3 dictation and meeting notes in a desktop app. Pick Nemotron 3.5 Super VL in ElevenAgents if you want a cheaper open LLM behind your voice agent. Interaction Engine, Memory, Routines and voice fine-tuning are all still on the roadmap, and I'll write them up when they land in the docs changelog.

—Anabel ♡