Voxtral Mini 3B 2507 (Amazon Bedrock, US)

Voxtral Mini 3B 2507 is an open-weights audio-language model from Mistral AI that extends the Ministral 3B text backbone with native speech understanding. Positioned as a compact 3-billion-parameter release in the Voxtral family, it targets use cases such as speech transcription, translation, and direct audio question answering, combining ASR-style capabilities with on-model reasoning rather than relying on separate transcription and language pipelines. The model is delivered through a public Hugging Face repository under the Mistral AI organization, signaling an open distribution model alongside its text and audio inputs.

In practical terms, the model is built for long-form and multilingual spoken interactions. It supports a 32k token context window that can absorb up to roughly thirty minutes of audio for transcription or forty minutes for understanding-style tasks, and it ships with a dedicated transcription mode that automatically detects the source language before producing text. A built-in Q&A and summarization path lets users ask questions directly over audio and generate structured summaries, while automatic language detection covers widely used languages including English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian. Function-calling from voice inputs further broadens its fit for assistant and workflow applications that need to act on spoken intents while preserving the text capabilities inherited from its Ministral lineage.