Mistral: Voxtral Small Transcribe 3

Mistral: Voxtral Small Transcribe 3 is a speech-to-text model from Mistral, listed on api14 under the transcription category. It is described as Mistral's flagship transcription model, and its provider release date is 20 October 2026. If your work involves turning recorded speech into written text, this is the model family to look at first from this provider.

The model accepts audio as its only input modality and returns a transcription as its output. According to the provider description, the transcript includes timestamps and speaker diarization, so you get both when something was said and an indication of who said it. It also supports context biasing, which lets you steer recognition toward domain-specific vocabulary. The context window is 65,536 tokens, and the maximum output is 52,428 tokens, which is a large ceiling for a single transcription job.

Those capabilities point to a fairly specific set of uses. Meeting and call recordings are an obvious fit, because timestamps and diarization make it practical to produce readable minutes or to jump to a particular speaker's remarks. Interview transcription benefits from the same two features, since you can separate interviewer from subject without manual splitting. Context biasing matters in technical, medical, legal or otherwise specialised settings, where names, acronyms and jargon are common and generic recognition tends to get them wrong. Any workflow that needs a searchable, time-aligned text version of an audio archive can be built on this model.

You call it through api14's OpenAI-compatible API using the model id "mistralai/voxtral-small-transcribe-3-0". Because the interface follows the OpenAI format, you can point an existing OpenAI client at api14 and send audio in the request rather than rewriting your integration. The model's supported parameters list is empty, so there are no extra tuning options to set beyond what the endpoint itself accepts; keep your request to the standard fields and let the model handle the transcription. The live price table on this page covers what you pay per use.

One thing to weigh is scope. This model is built for transcription only, with audio in and text out, so it is not the right choice for chat, reasoning, image or general text generation tasks; pick a general purpose or multimodal model for those. It also does not expose tunable parameters, so if you need fine control over decoding behaviour, check whether another transcription model on api14 offers it. For straightforward speech-to-text with timestamps, diarization and vocabulary biasing, it is a direct fit.