Voxtral Mini 3B
A solid pick for self-hosting and full data control.
Overview
Voxtral Mini 3B is Mistral AI's audio-enabled extension of Ministral 3B, built to transcribe, translate and summarise speech across eight languages within a 32k-token window that covers up to 40 minutes of audio per pass. Its real limitation: system prompts aren't supported yet, which complicates integration into agentic pipelines. Released under Apache 2.0 and light enough to run on a laptop, it keeps all data on-premise — a genuine plus for European teams prioritising data sovereignty.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Confidential meeting transcription
- Multilingual voice translation
- Automated call summarization
- Voice-triggered API calls
- On-prem subtitle generation
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 1.7 GB | 4 GB | Any recent PC/Mac |
| Q8 · balanced | 3.2 GB | 6 GB | Any recent PC/Mac |
| FP16 · max quality | 6 GB | 9 GB | 16GB PC / M1+ Mac / 8GB GPU |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve mistralai/Voxtral-Mini-3B-2507 \
--max-model-len 32000 \
--tensor-parallel-size 1 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |