Voxtral Small is a multimodal speech-and-text model from Mistral AI that brings audio understanding to a compact, locally runnable footprint. Built as an enhancement of Mistral Small 3, it adds state-of-the-art audio input capabilities while retaining strong text performance. The model can operate in a dedicated transcription mode, supports voice-driven function calling, and includes built-in Q&A and summarization without needing separate speech-to-text and language model pipelines. It natively handles multilingual audio, automatically detecting language and transcribing or translating across a broad set of spoken languages.

The Voxtral family was trained to comprehend both spoken audio and text documents, achieving state-of-the-art performance across a range of audio benchmarks while outperforming several closed-source models. Voxtral Small handles long-form audio content suitable for extended interviews, meetings, or podcasts, and is released with open weights under the Apache 2.0 license. This makes it a practical choice for teams that want to run a capable speech model on their own infrastructure while still having access to managed cloud inference through platforms like Amazon Bedrock.