ElevenLabs released v4 and v4 Turbo, new text-to-speech models that support 90 languages and can clone a voice from 10 seconds of audio, TechCrunch reported Monday. The company is aiming the release at businesses building voice agents, the fastest-growing part of its revenue.

What Happened

Language coverage rises to 90 from 70 in the v3 generation, with the biggest quality gains in Japanese, Brazilian Portuguese, Mandarin and Cantonese, according to TechCrunch. Voice cloning now works from a 10-second sample. ElevenLabs says on its product page that cloned voices require verified consent.

The models run on a new architecture that the company says reads a script the way a voice actor would, using the surrounding text to judge who is speaking and how a line should be delivered. ElevenLabs says this also keeps a speaker's voice consistent when lines are regenerated, reducing the drift that can creep into long narration or dialogue.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

Users can now stack several inline expression tags, such as cues for laughing or whispering, on the same passage. The company says v4 follows sequences of tags, including sound effects, more reliably than v3.

For conversational use, lower latency lets a voice agent begin speaking while the underlying language model is still streaming its response. ElevenLabs lists median inference latency for v4 Turbo at about 100 milliseconds. TechCrunch reported that the models are also better at distinguishing confrontations, escalations and holds, situations common in customer service calls.

Why It Matters

Voice agents for call centers, sales and support have become the main commercial use of synthetic speech, and those buyers care about latency and consistency more than about novelty voices. A model that can start talking before a chatbot has finished generating text removes one of the pauses that makes automated calls feel robotic.

Broader language coverage matters for the same customers. A company running one agent platform across several countries needs comparable voice quality in each market, and the improvements ElevenLabs highlighted in Japanese, Mandarin, Cantonese and Brazilian Portuguese cover some of the largest consumer markets outside English-speaking countries.

The business numbers show why ElevenLabs is leaning into enterprise work. TechCrunch reported that the company's annualized revenue run rate has climbed to more than $600 million from about $330 million at the start of the year, with more than 55% of the business coming from large enterprises. Headcount is above 800.

Competition and Funding

ElevenLabs competes with speech specialists including Cartesia, Deepgram and Fish Audio, as well as with Google and OpenAI, which offer voice capabilities alongside their language models. On its v4 page, the company compares its latency favorably with Cartesia and OpenAI, figures that come from ElevenLabs itself.

The company raised $500 million from Sequoia at an $11 billion valuation in February and is reportedly raising again at $22 billion. Chief executive Mati Staniszewski told TechCrunch the company intends to go public in the coming years but did not give a timeline.

What to Watch

The v4 models are available on all ElevenLabs plans, including the free tier, according to the company. The test will be how quickly enterprise customers move production voice agents onto v4 and whether independent evaluations confirm the latency and multilingual gains. A new funding round at the reported valuation, and any concrete IPO preparations, would be the next financial milestones.