Mistral: Voxtral Mini Transcribe 3

Mistral Voxtral Mini Transcribe 3 is a speech-to-text model from Mistral, released by the provider on 20 October 2026. It sits in the transcription category and is the compact member of Mistral's Voxtral transcription line, which makes it the option to consider when you want Mistral's speech recognition without a larger model. Its model id on api14 is mistralai/voxtral-mini-transcribe-3-0.

The model accepts audio as its only input modality and returns a transcript as its only output. Two features stand out in the transcript itself: timestamps and speaker diarization, so the output can tell you when something was said and who said it. It also supports context biasing for domain terms, which lets you steer recognition toward vocabulary such as product names, jargon or proper nouns that a general model might mishear. The context window is 65,536 tokens, and the maximum output is 52,428 tokens.

Those capabilities point to a fairly specific set of jobs. Meeting and call transcription is the obvious one, since diarization separates participants and timestamps let you align the transcript with a recording. Interview transcription follows the same pattern. Context biasing makes the model a better fit for specialized audio, for example clinical notes, legal proceedings, technical talks or customer support calls where in-house terminology appears often. Because the output is a transcript rather than a summary or an answer, you would typically pass the result to a separate text model if you need summarization, extraction or search indexing downstream.

You call the model through api14's OpenAI-compatible API using the model id mistralai/voxtral-mini-transcribe-3-0, sending audio as the input. The model lists no supported parameters, so there are no sampling or decoding knobs to tune on the request; keep your calls to the standard transcription shape and rely on the model's defaults. If you need to influence recognition of specific terms, that is what context biasing is for rather than a request parameter.

The main thing to weigh is scope. This is a transcription model with audio input and transcript output only, so it will not answer questions about the audio, translate in a separate step or generate text beyond the transcript. If your workload is text rather than speech, or you need a model that reasons over the content it receives, a general text model is the better choice. Within speech-to-text work, Voxtral Mini Transcribe 3 fits best when you want timestamps, diarization and domain term biasing in a compact Mistral model.