ElevenAI Is Live: ElevenLabs Beyond Voice Generation

ElevenLabs launched ElevenAI, a free early-access desktop app for meeting transcription, voice dictation in 90+ languages, and AI-driven follow-up actions, at its first summit in New York on November 11. The app runs on Scribe v3, a speech-to-text model launched November 10 that cuts word error rates 27% versus Scribe v2 at the same $0.20 per hour and ranks second among 64 models on Artificial Analysis with a 1.9% word error rate; Scribe v3 Live adds real-time speaker diarization with a 43% error reduction. ElevenAI has no public developer API yet, so builders work through the ElevenAgents platform, which previewed Memory for persistent context and Routines for multi-step automation, neither with a GA date.

ElevenLabs built its name on voice. Text-to-speech for podcasters, voice cloning for creators, real-time synthesis for game developers. That was the product. That was the pitch. On November 11, the company held its first-ever summit in New York and walked out with a very different identity. The marquee announcement was ElevenAI — a workplace app for meeting transcription, voice dictation, and AI-driven follow-up actions. ElevenLabs is not pivoting. It is expanding. The question for developers is what that expansion means for the platform they are already building on. What ElevenAI Does ElevenAI is a desktop app, currently in free early access. You have to sign up for access rather than downloading it directly. The core feature set covers three things: meeting notes, dictation, and agent automation. The meeting side transcribes spoken conversations — in-person or virtual — with speaker identification and automatic action item extraction. It sends each attendee a summary after the call. Critically, it only listens with everyone’s consent, which matters for enterprise compliance. During a meeting, it can surface relevant context live, not just after the fact. Dictation lets you speak into any app and get cleaned-up text, in any of 90+ languages, with support for industry-specific terminology. That language coverage is wider than any competitor in this space — Otter.ai supports six languages, Fireflies supports 60. ElevenAI is out in front on this metric. The third piece is where it gets interesting for developers: ElevenAgent integration. Companies can connect ElevenAgents to their ElevenAI setup to trigger automated follow-up workflows from meeting content. The straightforward example is drafting a follow-up email from your notes. The more interesting cases involve routing information from a conversation directly into downstream systems. The Underlying Model: Scribe v3 Both meeting and dictation features run on Scribe v3 https://elevenlabs.io/blog/scribe-v3 , ElevenLabs’ new speech-to-text model that launched alongside ElevenAI on November 10. Scribe v3 cuts word error rates by 27% compared to Scribe v2, while keeping the price exactly the same at $0.20 per hour. On the Artificial Analysis benchmark board https://artificialanalysis.ai/speech-to-text tracking 64 models, it currently ranks second with a 1.9% word error rate. If you are using Scribe v2 anywhere in your stack today, switching to v3 is a straightforward upgrade. Same endpoint, same pricing, meaningful accuracy improvement. ElevenLabs also released Scribe v3 Live for real-time transcription, which brings speaker diarization to real-time for the first time — a previous limitation — with a 43% reduction in errors versus the previous version. A domain-specific variant, Scribe v3 Medical, is already available inside ElevenAgents for healthcare terminology. A financial services variant is on the roadmap. What Developers Should Pay Attention To ElevenAI the app does not have a public developer API yet. If you want to build on it, you are working through ElevenAgents — the existing platform. That is not a limitation so much as the current stage of the product. What ElevenLabs previewed at the summit points toward where the platform is heading. Memory gives ElevenAgents persistent context across sessions, so a returning user does not have to re-explain their situation. Routines add a multi-step automation layer that runs before, during, and after conversations — handling follow-up work like callbacks, appointment management, or lead processing. Neither Memory nor Routines has a GA date yet, but both were announced with enough specificity to be worth factoring into architectural decisions you are making now. The ElevenAgents API https://elevenlabs.io/docs/eleven-agents/overview recently gained noisy environment detection and reusable evaluation criteria. The team is shipping developer-facing improvements on a weekly cadence, not waiting for the next summit. The Competitive Picture ElevenAI enters a crowded market. Otter.ai owns the transcript library experience with the strongest search functionality for English. Fireflies.ai https://fireflies.ai leads on CRM sync — Salesforce, HubSpot, and Pipedrive are native integrations. Notion AI is the natural choice for teams already living in Notion. ElevenAI’s structural advantage is language breadth and agent-native architecture. No other meeting tool has 90+ language support, and none are built on top of a voice AI platform that also handles synthesis, cloning, and real-time agent conversations. The gap to close is CRM integration — that is where Fireflies retains a clear advantage for sales-focused teams. The Summit Context ElevenLabs launched 13 products on November 11. Beyond ElevenAI, the most significant announcements were an Interaction Engine that routes conversations to the cheapest capable model mid-session, custom domain-specific STT fine-tuning on enterprise data, and a Partner Program now spanning 550 companies across 44 countries — including Google Cloud, IBM, Deloitte, and McKinsey. The company also expanded its Nvidia compute deal by 15 MW for large customers, and completed an STT/TTS integration with IBM watsonx Orchestrate. The company is not trying to stay a voice model vendor. It is building voice infrastructure for the enterprise stack. ElevenAI is the most visible part of that strategy — but the custom models and agent platform underneath it are what will determine whether this expansion sticks. Sign up for ElevenAI early access https://elevenlabs.io directly from ElevenLabs. If you have not updated your Scribe integration to v3 yet, the accuracy is better and the bill is the same — there is no reason to wait.