ElevenLabs

Origins and founders

Staniszewski and Dąbkowski grew up in Poland. Poor film voiceovers prompted their interest in speech that could retain a speaker’s identity across languages. Their June 2023 announcement dates the company’s start to April 2022 and its public beta to January 2023.

The founders brought different backgrounds. Dąbkowski had worked in machine learning at Google. Staniszewski had worked at Palantir. The original product focused on generating spoken performances from text.

Research and product development

The company organized its services around ElevenCreative for media production, ElevenAgents for conversational applications and ElevenAPI for developers. Its research now spans speech generation, transcription, sound effects, dubbing and music.

Voice Library also became a marketplace for verified professional voice replicas. A creator must complete voice verification before sharing a professional clone. The creator can control availability and receive payments when eligible subscribers use it. These voice profiles are distinct from the underlying speech models.

Eleven Realtime Mini and Max configure recognition, turn-taking, an LLM and speech generation. They are integrated agent offerings, not verified additional foundation models.

Ownership and growth

The February 2026 Series D was led by Sequoia, and Andrew Reed joined the board. Andreessen Horowitz and ICONIQ also increased their investment. The announcement described continued expansion in cities that include London, New York, Warsaw and Tokyo.

In September 2026, ElevenLabs said it had closed an employee tender led by Wellington and T. Rowe Price. This was a transaction in employee shares. Its valuation is distinct from the amount of new funding raised in a financing round.

Curated music adaptations

In March 2026, Music Finetunes added custom training from uploaded tracks and 11 curated adaptations: Afro House, Reggaeton, Arabic Groove, 70s Cambodian Rock, 80s Nu-Disco Revival, 1970s Rock Francais, Golden Hour Indie Guitar, Wooden Slit Drum, Brazilian Funk, Baile Beats and Mozart-Style Symphony.

The announcement names these as trained adaptations but does not supply checkpoint identifiers, individual release specifications or an explicit base-version mapping. They are kept together as a music-family feature rather than assigned invented standalone model identities.

Models and access

These are hosted model services. The service terms distinguish rights to generated output from rights to the underlying models. Permission to download generated audio does not provide a license to the model weights.

The original v1 speech models were scheduled for removal on 15 December 2025. Music v1 and Sound Effects v2 are deprecated, with later removal dates.

Voice Remixing uses the Voice Design v3 model. Generated voice IDs, audio formats and saved clones are not separate foundation-model releases. Dubbing v1 remains a product option, while Dubbing v2 has explicit documentation of a new audio-to-audio architecture.

Published latency figures depend on their measurement method. ElevenLabs reports median server measurements and excludes network and application time. Its speech-generation measurements use warm voices; recognition latency measures time until a committed transcript.

Realtime Translation is an Enterprise alpha service that uses v4 Turbo by default. The reviewed announcement does not establish a separately trained translation model. The music service also offers curated fine-tunes and customer-created adaptations.