Voxtral Mini 3B

Voxtral Mini 3B is a compact multimodal audio model created by Mistral as part of its first open-source audio family, sitting alongside the larger Voxtral Small. It is built on the Mistral 3B architecture, extending an existing text-capable backbone with state-of-the-art audio input handling while preserving strong text performance. The model's design emphasizes real-time audio understanding, making it well suited to workflows where spoken content needs to be converted into structured written output rather than processed as raw text alone.

The model targets practical audio-to-text applications, including speech transcription, translation, summarization, and question answering over spoken input, and its efficiency makes it a good fit for offline and edge-style deployments where larger speech models would be too costly. It supports long-context reasoning, which helps when transcribing or analyzing extended recordings such as meetings, lectures, or podcasts, and it pairs naturally with high-throughput inference frameworks like vLLM for building responsive audio assistants. Because the weights are openly available, developers can fine-tune or self-host the model for domain-specific voice tasks without depending on a proprietary speech stack.

Privatemode AIvoxtral-mini-3bvoxtral

Open provider page Provider docs

Quick Info

Provider: Privatemode AI
Model key: voxtral-mini-3b
Release date: Jul 1, 2025
Last updated: Jul 1, 2025
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 32,000 tokens
Context window: 32,000 tokens

Latest news about Voxtral Mini 3B

Privatemode AI

CoverageBenchmark

The Voxtral family tracker page explicitly lists Voxtral Mini 3B (2507) as a 3-billion-parameter audio model with a composite score of BB65.2, released in July 2025, available as open weights (with no API-only variants in the family). It confirms the (2507) version tag and notes the model's primary modality as audio, a The page also reports aggregate family metrics including 566K total Hugging Face downloads across the two checkpoints and 100% of family models graded B. The entry is a third-party aggregated tracker rather than a first-party Mistral or Privatemode AI release note, so it restates known specifications without providing

Privatemode AI

Coverage

Voxtral Mini 3B is documented as Mistral AI's 3-billion-parameter speech-understanding checkpoint in the open-weight Voxtral family, released in July 2025 under the Apache 2.0 license and positioned for local and edge deployments alongside the larger Voxtral Small 24B. The glossary describes Voxtral Mini 3B as supporti The same source notes automatic language detection across French, English, Spanish, Portuguese, Hindi, German, Dutch, Italian, and Arabic, and states that the model retains the text capabilities of its Mistral Small 3.1 backbone, eliminating the need to chain a separate ASR system with a language model. While useful as