英国AI语音公司ElevenLabs在2026年第三季度至第四季度期间,通过完成3亿美元员工老股转让将估值翻倍至220亿美元,并密集发布Eleven v4、Realtime及Instant Soundtracks等模型以巩固其“AI语音层”地位。【背景】尽管面临客户如Decagon反向竞争及模型质量趋同压力,CEO Mati Staniszewski明确表示愿牺牲毛利率以换取市场份额,强调企业级部署中情感智能与低延迟交互的关键作用。当前争议在于:随着音频模型逐渐商品化,ElevenLabs能否在保持技术领先的同时,应对来自应用层客户的直接竞争及开源模型的冲击。
Como chegamos aqui
CEO Mati Staniszewski在Nrth会议表示不介意压缩毛利率以扩大市场份额,披露ARR达6亿美元,55%以上来自企业客户。
ElevenLabs has introduced the Eleven Realtime series of models, which integrate speech recognition, LLMs, and voice generation into a single system to simplify the development of natural voice agents. The official video indicates that both v1 Mini and v1 Max models can begin responding in approximately 0.6 seconds, with v1 Max scoring 98.8% on Artificial Analysis's Big Bench Audio test. The model is available via the ElevenAgents platform and an API compatible with OpenAI's Realtime API.
In a video introduction, ElevenLabs presented its new Speech-to-Speech model series, Eleven Realtime. The models integrate speech recognition, turn-taking, LLM, and voice generation into a single system to simplify the development of natural voice agents. Official claims state that the v1 Max version scored 98.8% on Artificial Analysis's Big Bench Audio test with an average time-to-speak of 0.57 seconds. Developers can access it via ElevenAgents or through an API compatible with OpenAI's Realtime API.
In September 2026, ElevenLabs co-founder and CEO Mati Staniszewski stated clearly at the Nrth conference in Toronto that the company does not mind further gross-margin compression if it means expanding market share. This came against the backdrop of customer Decagon having trained its own voice product on ElevenLabs and now competing with it, revealing the actual economic strategy of an AI voice infrastructure company under pressure toward model quality convergence.
At the 2026 Bengaluru Summit, ElevenLabs India Head Karthik Rajaram and engineer Adarsh Shiragannavar showcased the scaling of voice AI in India. The session featured a live demonstration of building an insurance claim agent using Architect, alongside discussions on platform infrastructure, data privacy, and localized deployment strategies.
ElevenLabs released its most expressive text-to-speech model, Eleven v4, alongside the low-latency Eleven v4 Turbo on September 30, 2026. The new models feature a novel architecture designed to interpret tone, pacing, and emotion for more performative speech generation, with significant improvements in multi-speaker dialogue naturalness. v4 Turbo targets real-time agents with approximately 100ms median latency. The update also enhances audio tag adherence and support for over 90 languages, with notable gains in Japanese, Brazilian Portuguese, and Chinese.
On September 29, 2026, ElevenLabs released a video introducing its latest voice model, Eleven v4 Turbo, within the ElevenAgents platform. The model highlights a median inference latency of approximately 100 milliseconds and enhanced emotional expressiveness, aiming to support sensitive real-time conversations in healthcare, finance, and government sectors. The video details improvements in multilingual support and long-call reliability, emphasizing co-optimization with transcription and turn-taking models.
On October 15, 2026, ElevenLabs released a video introducing Instant Soundtracks. This feature allows users to upload footage, after which the model generates original music and places sound effects at appropriate timestamps. It supports using original audio as a reference to match the new score's character. The update also improves timing for short-form content and mood switching accuracy across scenes. In the Music Video Editor, sound effects appear as separate layers for independent adjustment. The feature is now available in ElevenCreative, ElevenMusic, and ElevenAPI.