ElevenLabs: Eleven Multilingual v2

Eleven Multilingual v2 is a text-to-speech model from ElevenLabs, offered through api14 under the model id elevenlabs/eleven-multilingual-v2. It sits in the speech category and is built for turning written text into spoken audio. ElevenLabs positions it for lifelike, consistent long-form narration, which tells you where it is meant to fit: sustained reading rather than short one-off clips.

The model takes text as its only input and returns speech as its only output. It supports 29 languages, so a single model can cover multilingual projects without swapping between separate voices or systems. The provider describes it as suited to long-form narration such as voice-overs and audiobooks, and it accepts requests of up to 10,000 characters. A set of named voices is available, including george, sarah, adam, alice, bella, bill, brian, callum, charlie, chris, daniel, eric, harry, jessica, laura, liam, lily, matilda, river, roger and will. No context length in tokens is published for this model, and the supported parameters list is empty, so plan around the character limit and the voice selection rather than sampling or token controls.

The practical uses follow directly from those traits. Audiobooks and long-form voice-overs are the clearest fit, since consistency across a long script is the stated strength. Multilingual narration for e-learning courses, product walkthroughs or documentary-style content is another natural case, because the same model handles all 29 languages. You can also use it for narrated articles, audio versions of documentation, or character dialogue where you assign different named voices to different speakers. Anything that needs a steady, natural reading voice over a long passage is a reasonable candidate.

Calling it through api14 works like any other model on the gateway: send a request to the OpenAI-compatible endpoint with the model field set to elevenlabs/eleven-multilingual-v2, and the response comes back as speech. Because the supported parameters list is empty, there are no documented tuning knobs such as temperature or top_p to set; the meaningful inputs are the text itself and the choice of voice. Check the live price table on the page for current rates before you commit to a workload.

One limit worth noting is the 10,000-character request ceiling, which means very long scripts need to be split into multiple requests, and consistency across those chunks becomes something you manage yourself. If your task is text generation, translation, transcription or anything other than producing speech, this model is not the right tool, since it only accepts text and only returns audio. For speech work within the character limit, it is a straightforward choice.