Mistral: Voxtral Mini 3B 2507

Mistral: Voxtral Mini 3B 2507 is a speech and audio understanding model from Mistral AI. It sits in the transcription category, and its provider release date is 2026-08-13. The name reflects its design: a compact 3B model in the Voxtral line, versioned 2507. If you are already working with Mistral models elsewhere in your stack, this one extends that family into audio processing rather than text generation.

The model takes audio as its input modality and produces transcription as its output modality. Mistral describes it as suited for transcription, translation, and compact audio processing workloads, so those are the tasks to plan around. The provider description does not state a context window, and no context length is listed for this model, so treat long-form audio handling as something to verify against your own workload rather than assume. The listed maximum output is 0, which means you should not rely on a documented output token budget when designing your pipeline.

The practical uses follow directly from those capabilities. Transcription is the primary one: turning recorded speech into text. Translation is the second, which in an audio model typically means working with speech in one language and producing text in another. The third is compact audio processing, a broad label that covers smaller or lighter audio tasks where a 3B model is a reasonable fit. Because the model is small, it suits workloads where you want audio understanding without a large model, such as batch transcription jobs, meeting or call recordings, and pipelines that feed transcribed text into later stages.

You call it through api14's OpenAI-compatible API using the model id "mistralai/voxtral-mini-3b-2507". Because the interface is OpenAI-compatible, you can point an existing OpenAI client at api14 and send audio input in the usual way, with transcription returned as the output. Note that no supported parameters are listed for this model, so do not assume tuning options such as temperature or top_p are accepted. Send the model id and your audio input, and check the live price table on the page for current rates.

On limits, the facts here are thin. There is no stated context length and a maximum output of 0, so if your work depends on very long audio or a guaranteed output budget, confirm behaviour before committing. If you need text generation rather than audio transcription, a text model in the catalogue will fit better. For transcription, translation, and compact audio processing, this model is the direct match.