ElevenLabs launches 13 products and upgrades

The announcement in context

ElevenLabs, founded in 2022, has built its reputation on synthetic voice technology, with its text-to-speech and voice-cloning products adopted across media, publishing, and customer service. The New York Summit announcements arrive at a moment of intensifying competition among voice AI providers, as enterprises increasingly deploy conversational agents in customer-facing roles and as speech-to-text accuracy becomes a differentiating factor in regulated industries such as healthcare and financial services. The scale of this release, 13 products and upgrades presented in a single event, reflects the company's attempt to address that entire stack at once rather than incrementally.

Three new voice models

The core of the release consists of three voice models shipped within the last month. Scribe v3 is the company's newest speech-to-text model, supporting transcription in 102 languages. According to ElevenLabs' own evaluation, Scribe v3 makes 27 percent fewer word errors than its predecessor, Scribe v2, across those 102 languages. That figure is the company's own testing and has not been independently verified. A healthcare-tuned variant, Scribe v3 Medical, has also launched, and Scribe v3 Live, a real-time transcription product, launched on 10 November 2026, one day before the Summit.

Eleven v4 is the company's newest expressive text-to-speech model, designed to produce speech with greater control over tone, pacing, and emotional register. Eleven Realtime, the third model, enables speech-to-speech conversational agents that listen and respond without an intermediate text stage, reducing latency in live interactions. The three models are positioned as complementary: Scribe v3 captures spoken input, Eleven Realtime sustains the conversation, and Eleven v4 renders the agent's spoken output.

ElevenAI enters free early access

The most prominent product announcement is ElevenAI, a workplace application that converts spoken meetings into written notes, summaries, and action items. ElevenAI opened in free early access on 11 November 2026. The distinction matters: the product is available for early testing at no cost, but it has not reached a general commercial launch, and its feature set and pricing for a full release remain undisclosed.

ElevenAI places ElevenLabs in direct competition with established meeting transcription and summarisation tools. Its stated advantage is the integration of the company's own speech models, including Scribe v3, with agentic capabilities that can identify decisions and follow-up tasks from a conversation rather than merely transcribing it. Whether that integration produces measurably better meeting records than incumbent tools will depend on evidence from real-world use, which is not yet available.

Expanded collaboration with NVIDIA

ElevenLabs also announced an expanded relationship with NVIDIA. The chipmaker's Nemotron 3.5 Super VL model is now available within ElevenAgents, the company's platform for building voice agents. In addition, ElevenLabs plans to add 15 megawatts of NVIDIA Vera Rubin compute capacity. That capacity is planned, not yet live, and no deployment date has been disclosed. The announcement signals growing infrastructure investment, but the compute expansion should be read as a stated intention rather than an operational fact.

A partner ecosystem across 44 countries

The company's Partner Program now counts more than 550 partners across 44 countries. ElevenLabs announced an alliance with McKinsey, working with the consultancy's QuantumBlack AI practice, and introduced a Marketplace through which more than 20 partners offer industry-specific applications built on ElevenLabs technology. Purchases through the Marketplace will be possible through customers' existing commitments only from 2027, a detail that limits its near-term commercial effect.

The partner figures describe the breadth of the ecosystem rather than its depth. ElevenLabs has not disclosed revenue attributable to partners, nor the number of partners actively deploying production systems, so the commercial significance of the 550-partner figure cannot be assessed from the announcement alone.

What remains in preview

Several announced capabilities are not yet available. The Interaction Engine, a system for more natural conversational behaviour, was previewed rather than released. Memory, Routines, and Architect, features intended to give agents persistent context and structured workflows, were likewise presented as forthcoming. Enterprise fine-tuning is limited to selected enterprises and is expected to arrive in the coming months, while a financial-services model remains in development only. These items should be understood as a product roadmap, not as shipping software.

Why the release matters

The announcement is significant for three reasons. First, the simultaneous release of recognition, synthesis, and real-time conversation models gives ElevenLabs a complete voice stack under one roof, reducing its dependence on third-party components. Second, ElevenAI moves the company from developer infrastructure into end-user workplace software, a larger and more contested market. Third, the NVIDIA collaboration and the McKinsey alliance suggest the company is positioning itself for enterprise deployments where infrastructure scale and consulting distribution matter as much as model quality.

At the same time, the release leaves important questions unanswered. The accuracy improvement claimed for Scribe v3 rests on the company's own evaluation, and no independent benchmark has been published. The 15-megawatt compute expansion has no timeline. The Marketplace will not support purchases through existing commitments until 2027. These gaps do not diminish the announcements, but they define the limits of what can presently be concluded about them.

What to watch next

The immediate test for ElevenLabs is whether ElevenAI gains traction during its free early access period and whether the company converts that usage into paying customers. A second test is the delivery schedule for the previewed capabilities, particularly enterprise fine-tuning and the financial-services model, where regulated customers will require evidence of accuracy and compliance before deployment. A third is whether the planned NVIDIA compute capacity materialises on a disclosed timeline. Each of these will provide a clearer measure of the release than the announcement itself.