ElevenLabs used its New York Summit on November 11, 2026 to show where the voice AI company wants to go next: past text to speech and into the full stack of customer and workplace conversations. The list is long, so here is what was announced, what is live today, and what is only a preview.
The short version: one new app is open for free early access (ElevenAI), new speech-to-text models shipped the day before (Scribe v3 and Scribe v3 Live), NVIDIA's Nemotron 3.5 Super VL is now available in ElevenAgents, and several bigger pieces (fine-tuning, Memory, Routines, Interaction Engine) are described as coming.
CEO Mati Staniszewski framed the day this way in the ElevenLabs Summit recap: "Within a few years, conversational agents will become critical infrastructure for every enterprise, as essential as cloud or payments."
What is ElevenAI?
ElevenAI is a new app that ElevenLabs showed as a sneak preview. The company calls it an AI teammate for enterprise teams. It listens to spoken conversations, including in-person meetings, and turns them into notes and next steps.
What ElevenLabs says it does:
- Dictation and meeting notes, powered by Scribe v3.
- In meetings, it takes notes with everyone's consent, then gives each attendee a summary and action items when the call ends.
- It can give live assistance during a call.
- Notice and consent mechanisms are built in.
Free early access opened on November 11.
What changes in ElevenAgents?
ElevenAgents is the company's platform for voice and chat agents in sales, support, recruitment and operations. Staniszewski shared several updates "coming to ElevenAgents."
Memory. An agent can recognize a returning, authenticated customer and recall what they said before. ElevenLabs says the CRM stays the primary customer record; Memory holds details that never had a CRM field. Its example: if a customer prefers Spanish and can only talk after 5pm, the agent remembers and acts on it.
Routines. This is a new automation layer for multi-step work before, during and after a conversation. ElevenLabs' examples include pre-screening a mortgage applicant and calling back within minutes, reminding patients before a visit, offering a canceled slot from the waitlist, emailing a report, writing to a CRM, or starting a call.
Interaction Engine. Separately, as part of its model work, ElevenLabs previewed a system that handles turn-taking and interruptions, so the agent knows when to speak and when to stop. It can also pick the LLM for each part of an exchange: a cheaper model for triage, and a frontier model for harder tasks.
Can businesses fine-tune ElevenLabs models?
Not yet. Staniszewski said that "in the coming months" ElevenLabs will help businesses fine-tune certain open-source language models, as well as ElevenLabs voice models. Businesses bring data from their live deployments and work with ElevenLabs researchers. Each business picks the metrics that matter, such as a customer satisfaction score (CSAT) or reliability. Staniszewski compared it to each company having its own reinforcement-learning stack.
ElevenLabs already uses a domain approach for Scribe v3 Medical, a healthcare transcription model. The company says it captures the difference between drug names like cycloserine and cyclosporine. A domain-optimized model for financial services is in the works.
If you follow ElevenLabs on the voice side, this sits next to recent model work like Eleven v4 professional voice clones.
How much better is Scribe v3?
Scribe v3 and Scribe v3 Live launched on November 10, the day before the Summit. These are ElevenLabs' own figures:
- Scribe v3 (recorded audio): 27% fewer word errors than Scribe v2 across 102 languages on an internal version of the FLEURS benchmark.
- Scribe v3 Live (realtime audio): 43% fewer errors than Scribe v2 Realtime. It adds speaker diarization (who said what) and improved turn detection.
- Speaker labels: Scribe v3 makes 17% fewer diarization errors than Scribe v2 on recordings with three or more speakers.
- Keyterm prompting: up to 1,000 terms, such as brand names and jargon.
Both models are in ElevenAPI, and Scribe v3 Live is available in ElevenAgents. ElevenLabs says switching from Scribe v2 is a one-line change to the model ID. For how other transcription tools compare, see our AI transcription tools comparison.
What is in the NVIDIA deal?
In a separate post, ElevenLabs and NVIDIA said they are expanding their partnership:
- Nemotron in ElevenAgents. NVIDIA Nemotron 3.5 Super VL is now available in ElevenAgents. ElevenLabs says that on its ElevenAgents benchmark for realistic customer-support tasks, the model achieved leading end-to-end task-completion accuracy at 70% lower costs, compared to other leading open models. With reasoning turned off, ElevenLabs says it is among the three fastest models on the platform.
- More compute. ElevenLabs plans to add 15 MW of NVIDIA Vera Rubin capacity, on top of its existing NVIDIA Blackwell Ultra systems, as part of expanding its on-premises infrastructure.
- Poland. ElevenLabs models and NVIDIA infrastructure power the voice agent for Centralna e-Rejestracja, the central appointment system in Poland's national e-Health platform, with EasyCall supporting delivery. The agent calls patients who only have landlines to remind them of appointments and help them reschedule. ElevenLabs cites 40 million appointments a year and no-show rates of 10 to 20 percent. It runs on-premises using NVIDIA Confidential Computing.
- Physical AI. The two companies are exploring on-device voice AI for machines, robots and vehicles. ElevenLabs describes these as potential future applications.
ElevenLabs notes that NVIDIA became a strategic investor in ElevenLabs in 2025.
Partner Program and Marketplace
ElevenLabs says its partner network has grown to more than 550 partners in 44 countries, including Google Cloud, IBM, Deloitte, DXC and Genesys. It launched the ElevenLabs Partner Program on November 11, plus an alliance with McKinsey and QuantumBlack, AI by McKinsey.
It also introduced the ElevenLabs Marketplace, a curated group of industry-specific AI and voice apps built on ElevenLabs by technology partners, for healthcare, financial services, insurance, education and retail. Starting next year, businesses will be able to buy these agents through the Marketplace using their existing ElevenLabs commitment.
What it means
Our read: very little of this is aimed at creators. Recent consumer moves, like ElevenMusic publishing to SoundCloud, sit on a separate track. The Summit was about enterprises: agents that remember customers, automate follow-up work and run on private infrastructure.
What you can use today is narrow: ElevenAI early access, Scribe v3 in the API, and Nemotron inside ElevenAgents. Memory, Routines, Interaction Engine and fine-tuning are announcements, not shipped features, and the accuracy and cost figures are ElevenLabs' own tests. The Poland deployment is the most concrete item on the list, because it is a working public service rather than a demo.
