Model · Mistral AI · Speech-To-Text
Voxtral Mini 4B Realtime
Mistral's open live-transcription model. It writes text as people speak, under half a second behind, in 13 languages — accuracy close to offline transcribers, small enough to run on one graphics card.
- Modality
- Speech-To-Text
- License
- Apache 2.0 (Open)
- Parameter size
- 4B (about 3.4B language model + 970M audio encoder)
- Released
- February 4, 2026
- Last verified
- October 28, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 8 GB (16 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 16 GB of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Mistral says it needs one GPU with 16 GB or more; Mac users run the 4-bit MLX build, and a MacBook demo is linked from the model card.
Numbers are for: official bf16 weights (GPU); 4-bit MLX build for Mac. Based on the maker’s own figures; where the maker gives only a graphics-card figure, the Mac memory is our estimate.
Strengths
- Apache 2.0 — free for commercial use
- Native streaming: about 480 ms delay matches leading offline open-source transcribers
- Delay is configurable, trading latency for accuracy
- Runs on a single GPU with 16 GB or more of memory
- Built for live subtitles, voice assistants and private meeting transcription
Weaknesses
- 13 languages — fewer than Whisper or Qwen3-ASR
- Long sessions need the context length set high (default about 3 hours)
- Mistral's newer Voxtral 3 transcription models are API-only previews, so the open model is a generation behind
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0, BF16; served with vLLM |
| Mistral API | hosted-api | Model id voxtral-mini-transcribe-realtime-2602; API key required |
Official sources
- Model card model
- Announcement announcement
- Technical report paper
- Mistral changelog docs
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602.
- — Initial entry. Added 2026-10-28 from the popularity sweep: 2.03M Hugging Face downloads/month. Released 2026-02-04 per Mistral's changelog (alongside Voxtral Mini Transcribe 2). Mistral released the Voxtral 3 audio family on 2026-10-20, including Voxtral Mini Transcribe Realtime 3, as API public previews; this open-weight 2602 model remains the current downloadable one.
Esc