Voxtral Mini 4B Realtime 2602
Voxtral Mini 4B Realtime 2602 is a 4B-parameter speech-to-text model from Mistral AI, released under the Apache-2.0 license. The reviewed source lists 13 languages (13 explicitly enumerated in our record), and the model is designed for native streaming. Among the common setups we test, it ranks highest for Japanese live transcription (#2 of 26 in the matcher).
Purpose-built 4B native-streaming ASR with configurable latency down to sub-200ms and 13 explicitly listed languages. Mistral releases the weights under Apache-2.0 and says the realtime path is not compatible with diarization. The official repository contains individual 8.86 GB weight files, so plan for capable local hardware and benchmark the exact latency setting you intend to use.
Key facts
| Provider | Mistral AI |
|---|---|
| Model family | Voxtral |
| Parameters | 4B |
| License | Apache-2.0 · Permissive — license text ↗ |
| Language coverage | 13 listed languages · 13 enumerated in our record |
| Streaming | Native streaming architecture |
| Timestamps | Limited or runtime-dependent |
| Speaker diarization | No — needs a separate component |
| Speech translation | No |
| Documented runtimes | vLLM (realtime API), Hugging Face Transformers, ExecuTorch — how to run it |
| Primary source | huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602 ↗ |
| Last reviewed |
Hardware and workload fit
These are Speech Model Match's conservative 1–5 editorial ratings, relative to the other 19 reviewed models. They come from model size, official memory guidance and documented runtimes — not from a universal benchmark. How the ratings work.
| Dimension | Rating | Scale |
|---|---|---|
| CPU-only fit | 1/5 | |
| Apple Silicon fit | 2/5 | |
| NVIDIA GPU fit | 5/5 | |
| Speed | 5/5 | |
| Accuracy (relative) | 5/5 | |
| Low memory use | 1/5 | |
| Long recordings | 3/5 | |
| Live / low latency | 5/5 | |
| Subtitle work | 5/5 | |
| Edge / low-power devices | 2/5 |
Where it is strong
- Native streaming architecture for live, low-latency transcription
- NVIDIA GPU fit: rated 5/5
- Speed: rated 5/5
- Accuracy (relative): rated 5/5
- Live / low latency: rated 5/5
- Subtitle work: rated 5/5
What to check first
- CPU-only fit: rated 1/5
- Apple Silicon fit: rated 2/5
- Low memory use: rated 1/5
- Edge / low-power devices: rated 2/5
- No native speaker diarization — pair it with a separate diarization pipeline if you need speaker labels
- Timestamp support is limited or runtime-dependent — test it if you need subtitles
How it ranks in common setups
Ranks come from the same deterministic matcher that runs on the homepage, applied to all 26 reviewed models. They are decision guidance, not accuracy claims.
| Setup | Rank | Fit score | Top match |
|---|---|---|---|
| Long recordings on an 8–12 GB NVIDIA GPU, balanced | #21 of 26 | 67/100 · Possible fit | Whisper turbo |
| Live captions or voice input (native streaming + timestamps, speed first) | #4 of 26 | 86/100 · Excellent fit | Nemotron 3.5 ASR Streaming 0.6B |
| Subtitles with timing, accuracy first | #11 of 26 | 77/100 · Strong fit | Whisper medium |
| Meetings with speaker labels (diarization + timestamps) | #17 of 26 | 67/100 · Possible fit | VibeVoice-ASR 7B |
| CPU-only machine, low memory use | #26 of 26 | 44/100 · Weak fit | VibeVoice-ASR-BitNet |
| Apple Silicon with 8–16 GB unified memory, long recordings | #24 of 26 | 57/100 · Weak fit | Whisper small |
| Japanese live transcription (native streaming, speed first) | #2 of 26 | 86/100 · Excellent fit | Nemotron 3.5 ASR Streaming 0.6B |
Supported languages
- Arabic
ar - Dutch
nl - English
en - French
fr - German
de - Hindi
hi - Italian
it - Japanese
ja - Korean
ko - Mandarin Chinese
zh - Portuguese
pt - Russian
ru - Spanish
es
Compare coverage across every reviewed model in the language support matrix.
Community activity on Hugging Face
- Downloads, last 30 days
- 1,539,259
- Likes
- 1,062
- Repository last updated
- 11 Mar 2026
- Published on the Hub
- 21 Jan 2026
Figures from the Hugging Face Hub API, 11 Nov 2026. Source: huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602 ↗. Downloads count file downloads, not active users, and do not affect the matcher's scores. Compare all models.
How to run Voxtral Mini 4B Realtime 2602 locally
Mistral recommends serving Voxtral Realtime with vLLM's realtime API over websockets. Runtimes below are the ones the official model card or repository documents for this model (checked November 2, 2026).
- vLLM (realtime API) ↗Serving engine · recommended by Mistral
- Hugging Face Transformers ↗Framework · listed from version 5.2; the card also calls vLLM the supported path
- ExecuTorch ↗Framework · on-device example, marked untested
Install command shown in the official huggingface.co page ↗:
uv pip install -U vllm
Audio input: Transcription delay is configurable in 80 ms steps; Mistral recommends 480 ms.
Official hardware guidance: Runs on a single GPU with at least 16 GB of memory (BF16 weights). Source ↗
Whatever runtime you choose, measure speed and accuracy on your own audio with the benchmark tool; results differ a lot between runtimes, precision and hardware.
Before you deploy Voxtral Mini 4B Realtime 2602
A printable checklist built from the facts on this page. Tick it off before putting the model into production.
- Language. Confirm your language is in the 13-language list on this page, and test any accents or code-switching you expect.
- License. Apache-2.0: Give recipients a copy of the license, keep copyright and attribution notices, carry forward any NOTICE file and mark files you changed. The patent license ends if you start patent litigation over the work. License guide.
- Install. Start from the official command:
uv pip install -U vllm(full run notes). - Audio input. Transcription delay is configurable in 80 ms steps; Mistral recommends 480 ms.
- Hardware. Runs on a single GPU with at least 16 GB of memory (BF16 weights).
- Features you rely on. For Voxtral Mini 4B Realtime 2602, native streaming is documented; timestamps are limited or runtime-dependent; speaker labels need a separate diarization component. Test each one you need with real audio.
- Measure. Run 10–20 representative clips, score them with the WER calculator and record speed in the benchmark workspace before you commit.
Alternatives and related models
Similar fit profile: Nemotron 3.5 ASR Streaming 0.6B (NVIDIA, 600M), VibeVoice-ASR Streaming 1.5B (Microsoft, 3B total), Qwen3-ASR 0.6B (Qwen, 0.6B). Similarity is computed from the ten fit ratings plus streaming and diarization support.
| Fact | Voxtral Mini 4B Realtime 2602 | Nemotron 3.5 ASR Streaming 0.6B | VibeVoice-ASR Streaming 1.5B |
|---|---|---|---|
| Provider · size | Mistral AI · 4B | NVIDIA · 600M | Microsoft · 3B total |
| License | Apache-2.0 Permissive | OpenMDW-1.1 Custom terms | MIT Permissive |
| Languages | 13 listed languages | 40 language-locales in tiers | 10 listed languages |
| Streaming | Native | Native | Native |
| Timestamps | Limited | Limited | Limited |
| Speaker diarization | Separate pipeline | Separate pipeline | Native |
| CPU · Apple · NVIDIA fit | 1/5 · 2/5 · 5/5 | 2/5 · 3/5 · 5/5 | 1/5 · 2/5 · 4/5 |
| Live · long recordings | 5/5 · 3/5 | 5/5 · 4/5 | 5/5 · 4/5 |
| Main runtime | vLLM (realtime API) | NVIDIA NeMo | microsoft/VibeVoice |
Head-to-head pages: Whisper turbo vs Voxtral Mini 4B Realtime 2602 · Nemotron 3.5 ASR Streaming 0.6B vs Voxtral Mini 4B Realtime 2602
Compare Voxtral Mini 4B Realtime 2602 side by side
Guides that cover Voxtral Mini 4B Realtime 2602
Check Voxtral Mini 4B Realtime 2602 for your setup
Pick your language, hardware and workload to see this model's fit score, the reasons behind it and where it ranks against all reviewed models. Everything runs in your browser.
Common questions
Can I use Voxtral Mini 4B Realtime 2602 commercially?
Permissive license; commercial use is allowed under the listed terms. The listed license is Apache-2.0; read the current license text ↗ before shipping a product, because terms can change. The license guide compares the obligations of all 26 reviewed models.
Does Voxtral Mini 4B Realtime 2602 support real-time streaming?
Yes. The reviewed source describes a native streaming architecture, so it is a candidate for live captions, dictation or voice agents. Measure end-to-end latency on your own hardware.
Which languages does Voxtral Mini 4B Realtime 2602 transcribe?
13 listed languages. The full list of 13 explicitly enumerated languages is shown above.
Can Voxtral Mini 4B Realtime 2602 tell speakers apart?
Not natively. Combine it with a dedicated diarization component and evaluate that component separately; the meetings guide covers this workflow.
Primary sources
Reviewed against primary sources on September 28, 2026. Model behavior, runtime support, language coverage and licenses can change; re-check the linked primary source before production use. See the full source directory.