ElevenLabs: Scribe v2

ElevenLabs: Scribe v2 is a speech-to-text model from ElevenLabs, a provider known for its work in audio and voice. It sits in the transcription category, so its job is converting spoken audio into written text rather than generating conversational replies. If your pipeline needs a dedicated transcription step, this is the kind of model you would evaluate on its own rather than as a general purpose chat model.

The model accepts audio as its input and returns a transcription as its output. According to the provider, it handles speech-to-text across more than 90 languages. It produces word-level timestamps, which let you align each word in the transcript to a point in the audio. Speaker diarization is available as an option, so you can distinguish who said what in a recording with multiple speakers. Audio-event tagging is also supported, which means non-speech sounds in the audio can be marked in the output. No context length is listed for this model, so treat it as a transcription endpoint rather than something you feed long text prompts into.

Those capabilities point to a fairly specific set of use cases. Word-level timestamps are useful when you need subtitles or captions that stay in sync with the audio, or when you want to jump to the exact moment a phrase was spoken. Speaker diarization fits meeting recordings, interviews, and call recordings where you need to attribute each line to a participant. Audio-event tagging helps when the transcript should note more than speech, for example marking a sound that occurs during the recording. The 90+ language coverage makes it a candidate for multilingual archives or products that serve users across regions.

You call the model through api14's OpenAI-compatible API using the model id "elevenlabs/scribe-v2". Because the interface follows the OpenAI format, you can point an existing OpenAI client at api14 and send your audio request with that id. The model lists no supported request parameters, so there are no extra tuning options documented here beyond what the endpoint itself accepts. The details also list a max output of 0, which is worth checking against your expected transcript lengths before you commit to it.

Given the empty parameter list and the absence of a stated context length, this model is best treated as a focused transcription service rather than a configurable text model. If you need a model that takes text prompts and generates text, or one where you can adjust sampling and similar settings, a different model on api14 will fit better. For audio in, transcript out, Scribe v2 is the direct option.