ElevenLabs

🤖 𝗪𝗲 𝗯𝗲𝗻𝗰𝗵𝗺𝗮𝗿𝗸𝗲𝗱 𝗘𝗹𝗲𝘃𝗲𝗻𝗔𝗴𝗲𝗻𝘁𝘀. 𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘄𝗲 𝗳𝗼𝘂𝗻𝗱. Building a great voice agent isn't just about a great model — it's about every layer of the stack working together under real conditions. That's why we recently released EVA-Bench, our open-source framework for evaluating voice agents end-to-end across 𝗘𝗩𝗔-𝗔 (Accuracy) and 𝗘𝗩𝗔-𝗫 (Experience). This week, we evaluated ElevenLabs’ ElevenAgents, and here are the results: • 𝗦𝗰𝗿𝗶𝗯𝗲 𝘃𝟮.𝟮 𝗥𝗲𝗮𝗹𝘁𝗶𝗺𝗲 𝗶𝘀 𝘁𝗵𝗲 𝘀𝘁𝗿𝗼𝗻𝗴𝗲𝘀𝘁 𝗦𝗧𝗧 𝗺𝗼𝗱𝗲𝗹 𝘄𝗲 𝘁𝗲𝘀𝘁𝗲𝗱. Transcription accuracy on key entities above 95%, holding above 93% under French accent and coffee shop background noise. • 𝗘𝗹𝗲𝘃𝗲𝗻 𝗙𝗹𝗮𝘀𝗵 𝘃𝟮 𝘀𝗰𝗼𝗿𝗲𝗱 𝗮𝗯𝗼𝘃𝗲 𝟵𝟳% 𝗼𝗻 𝗦𝗽𝗲𝗲𝗰𝗵 𝗙𝗶𝗱𝗲𝗹𝗶𝘁𝘆 - the only metric in any end-to-end voice agent benchmark that evaluates what the agent actually says out loud. • Of all 16 systems we benchmarked, 𝗯𝗼𝘁𝗵 𝗘𝗹𝗲𝘃𝗲𝗻𝗔𝗴𝗲𝗻𝘁𝘀 𝗰𝗼𝗻𝗳𝗶𝗴𝘂𝗿𝗮𝘁𝗶𝗼𝗻𝘀 𝗹𝗮𝗻𝗱𝗲𝗱 𝗼𝗻 𝘁𝗵𝗲 𝗣𝗮𝗿𝗲𝘁𝗼 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿. With Claude Haiku, the highest EVA-X score. With GPT-5.4, the highest EVA-A. Shaheen Lavie-Rouse, FDE at ElevenLabs, said it best: "𝘌𝘝𝘈-𝘉𝘦𝘯𝘤𝘩 𝘪𝘴 𝘵𝘩𝘦 𝘧𝘪𝘳𝘴𝘵 𝘣𝘦𝘯𝘤𝘩𝘮𝘢𝘳𝘬 𝘵𝘩𝘢𝘵 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 𝘮𝘦𝘢𝘴𝘶𝘳𝘦𝘴 𝘷𝘰𝘪𝘤𝘦 𝘢𝘨𝘦𝘯𝘵 𝘲𝘶𝘢𝘭𝘪𝘵𝘺 𝘦𝘯𝘥-𝘵𝘰-𝘦𝘯𝘥 - 𝘧𝘳𝘰𝘮 𝘵𝘳𝘢𝘯𝘴𝘤𝘳𝘪𝘱𝘵𝘪𝘰𝘯 𝘵𝘰 𝘵𝘢𝘴𝘬 𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘪𝘰𝘯 𝘵𝘰 𝘴𝘱𝘰𝘬𝘦𝘯 𝘰𝘶𝘵𝘱𝘶𝘵 - 𝘪𝘯 𝘢 𝘸𝘢𝘺 𝘵𝘩𝘢𝘵'𝘴 𝘢𝘤𝘵𝘪𝘰𝘯𝘢𝘣𝘭𝘦 𝘧𝘰𝘳 𝘱𝘦𝘰𝘱𝘭𝘦 𝘣𝘶𝘪𝘭𝘥𝘪𝘯𝘨 𝘢 𝘷𝘰𝘪𝘤𝘦 𝘢𝘨𝘦𝘯𝘵. 𝘞𝘦'𝘳𝘦 𝘢𝘭𝘴𝘰 𝘨𝘭𝘢𝘥 𝘵𝘰 𝘴𝘦𝘦 𝘵𝘸𝘰 𝘥𝘪𝘧𝘧𝘦𝘳𝘦𝘯𝘵 𝘌𝘭𝘦𝘷𝘦𝘯𝘈𝘨𝘦𝘯𝘵𝘴 𝘤𝘰𝘯𝘧𝘪𝘨𝘶𝘳𝘢𝘵𝘪𝘰𝘯𝘴 𝘰𝘯 𝘵𝘩𝘦 𝘗𝘢𝘳𝘦𝘵𝘰 𝘧𝘳𝘰𝘯𝘵𝘪𝘦𝘳." This is one of many evaluations we'll be releasing. If you're building voice agents — or building the models that power them — run EVA-Bench on your stack and show us where you land. 🔎 Want to know more? 🌐 𝗪𝗲𝗯𝘀𝗶𝘁𝗲: https://lnkd.in/eaRnvm7G 💻 𝗖𝗼𝗱𝗲: https://lnkd.in/e53m3GYe 🗂️ 𝗗𝗮𝘁𝗮𝘀𝗲𝘁: https://lnkd.in/erCqPkc6 📄 𝗣𝗮𝗽𝗲𝗿: https://lnkd.in/e_ShvWDw Technical Contributors: Tara Bogavelli, Gabrielle Gauthier Melançon, Katrina Stankiewicz, Oluwanifemi Bamgbose, Fanny Riols, Hoang Nguyen, Raghav Mehndiratta, Lindsay Brin, Joseph Marinier, Hari Subramani Leadership & Product: Sridhar Krishna Nemala, Anil Kumar Madamala, Srinivas Sunkara, Joyce Li, Nitin Aggarwal #VoiceAI #VoiceAgents #AIResearch #ConversationalAI #ServiceNowResearch #ElevenLabs

Introducing Dubbing v2, our revolutionary new dubbing model. For the first time, the performance of the original content carries over into every language, not just the words. Unlike other models, Dubbing v2 conditions directly on the original audio, preserving tone, emotion, and delivery. Sync-aware translation logic means phrasing adapts naturally, with starts and stops aligned to the original. Professional dubbing runs at hundreds of dollars per minute. Dubbing v2 brings that quality level to creators, marketers, and studios, fully automated, with no pipeline to build. Whether you're dubbing YouTube videos, localizing ads for every market, or producing broadcast-quality dubs through ElevenProductions, the performance lands the same way in 90+ languages. Available today in ElevenCreative and through ElevenProductions. Coming soon via ElevenAPI.

ElevenLabs is partnering with the Greek Government to reimagine public services. Today, our cofounder Mati Staniszewski signed an MOU with Greece’s PM Kyriakos Mitsotakis and Minister for Digital Governance Dimitris Papastergiou to use voice AI to improve public services, promote tourism, and preserve Greek cultural and linguistic heritage. Read more on the blog linked in comments.

Stan Lee is now on ElevenLabs. His voice replica is available on the Iconic Marketplace and in ElevenReader. Crafted from professional recordings, Stan’s wit and warmth comes through exactly as you remember it. This partnership with Stan Lee Universe goes further than voice: → The Stan Lee Book of the Month Club launches inside ElevenReader, starting with June's pick: Treasure Island by Robert Louis Stevenson. → His likeness is coming to the ElevenCreative Templates, so fans can generate cameo moments in the spirit of the appearances that defined his legend. → Two Stan Lee-inspired Finetunes are live on ElevenCreative Music, available to every ElevenLabs user. As the Stan Lee Universe team put it: "Fans have always told us that when they read his comics, they hear the words in Stan's voice. Now, thanks to ElevenLabs, we can make that a reality." Read the full announcement on our blog. Link in comments.

Today we're launching Music v2. Better vocals, instrumentation, and arrangement across every genre. Fast rap, dense lyrical delivery, mid-track genre transitions and non-musical sound effects embedded directly in a track. Plus improved multilingual generation and support for long-form composition, build a full song section by section. Music v2 powers ElevenMusic, ElevenAPI (coming soon), and ElevenCreative, so whether you're an artist, a developer, or a brand, there's a purpose-built platform for you. Plus, alongside this release we've cut Music pricing for ElevenCreative and ElevenAPI self-serve customers by up to 50%. Music v2 is trained only on licensed data and cleared for commercial use.

We’re excited to announce our next speaker for the ElevenLabs Summit in Warsaw: Michal Fijol, President of the Management Board and CEO at LOT Polish Airlines, the world’s 12th oldest airline and one of the most internationally recognized Polish brands.

Voice creators on ElevenLabs have now earned $22 million. This is up from $11 million in November, with earnings doubling in six months and the growth still accelerating. Over 10,000 creators across dozens of languages are earning on the Voice Marketplace. They upload a Professional Voice Clone, set their licensing terms, and earn every time someone generates with their voice. The catalog reflects the breadth of the community behind it - late night announcers, rugged cowboys, sinister villains, quirky mad scientists, warm narrators, and animated cartoon characters. For creators and marketers, the marketplace is an instant casting library: find the right voice for an audiobook, a product demo, a game character, or a campaign in any language. This is a new economic model for voice IP. The old model was one-time licensing fees - record a session, deliver the files, get paid once. The Voice Marketplace replaces that with ongoing earnings from every generation, on terms the creator sets, with consent and revocation built into the platform. We have extended the model to the Music Marketplace, which launched in March, and we are exploring ways to expand it further.

On June 1, we’re hosting the ElevenLabs Summit in Warsaw - bringing together leaders who are transforming how we interact with technology. We’re excited to announce our next speakers: Beata Podleśna-Gumkowska is Consumer AI Director at InPost, where she creates new AI-powered shopping experiences for customers and business partners. Tomasz Pol is VP of the Management Board at PKO Bank Polski, the largest bank in Poland, which serves 12.5 million customers.

We've added 200,000+ premium titles to ElevenReader, including best sellers from the world's leading publishers. For just $11/month, ElevenReader includes 20 hours of premium titles available in more than a dozen languages. Plus, unlimited written text as audio for your PDFs, articles, and docs. For a selection of titles in the app, listeners can also choose their narrator voice and personalize their listening experience.