Best larger-system route
Team 32
The smallest system in the range that reaches this model's preferred working allowance.
Explore Team 32Mistral AI model
Voxtral Realtime Hardware Requirements & Compatibility
The official BF16 files total about 17.7GB. One 32GB GPU is the first product fit, subject to the required vLLM audio support, stream count and latency target.
- Model version
mistralai/Voxtral-Mini-4B-Realtime-2602- Family and variant
- Voxtral Mini 4B Realtime · Voxtral-Mini-4B-Realtime-2602
- Source version
2769294da956- Source updated
- 11 March 2026
Mistral AI Voxtral Mini 4B Realtime
mistralai/Voxtral-Mini-4B-Realtime-2602 - Minimum GPU memory
- 24GB GPU planning floor
- Recommended hardware
- One 32GB GPU
- Licence
- Apache License 2.0
- Useful for
- Speech & audio
Buyer verdict
Where Voxtral Mini 4B Realtime is a sensible fit
Voxtral Realtime is relevant when live transcription matters more than offline batch speed and the team wants an Apache-licensed, current streaming model.
These figures apply to the named model version. Quantisation, fine-tuning, context length, image resolution, batch size and serving software can materially change the hardware needed.
Hardware requirements
- Minimum
- 24GB GPU planning floor
- Recommended
- One 32GB GPU
The 17.7GB weight files leave modest room for streaming buffers and runtime.
Concurrent stream capacity must be measured.
Technical specification
Voxtral Mini 4B Realtime model and hardware facts
Specifications shown for source version
2769294da9567371363522aac9bbcfdd19447add, updated
11 March 2026.
- Parameters
- About 4.4B
- Task
- Real-time automatic speech recognition
- Repository weights
- 17.72GB BF16
- Primary runtime
- vLLM
- Release
- February 2026 series
- Licence
- Apache 2.0
Product compatibility
Voxtral Mini 4B Realtime compatibility across all 11 GPU systems
- Systems that fit without model splitting
- 11
- Systems needing multi-GPU validation
- 0
A system is listed as fitting when its GPU memory meets the requirement shown above. Speed, usable context, batch size and concurrent users still need testing with the final model and software configuration.
sme workstation
Team 32
- Per GPU
- 32GB
- Total GPU memory
- 32GB
- GPU count
- 1
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Company 64
- Per GPU
- 32GB
- Total GPU memory
- 64GB
- GPU count
- 2
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Studio 96
- Per GPU
- 96GB
- Total GPU memory
- 96GB
- GPU count
- 1
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Studio 192
- Per GPU
- 96GB
- Total GPU memory
- 192GB
- GPU count
- 2
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Value Rack 128
- Per GPU
- 32GB
- Total GPU memory
- 128GB
- GPU count
- 4
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Value Rack 256
- Per GPU
- 32GB
- Total GPU memory
- 256GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Enterprise 384
- Per GPU
- 96GB
- Total GPU memory
- 384GB
- GPU count
- 4
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Enterprise 768
- Per GPU
- 96GB
- Total GPU memory
- 768GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
H200 1.1TB
- Per GPU
- 141GB
- Total GPU memory
- 1,128GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
frontier partner
Frontier Native 2.3TB
- Per GPU
- 288GB
- Total GPU memory
- 2,304GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
frontier partner
Frontier Rack 20TB
- Per GPU
- 288GB
- Total GPU memory
- 20,736GB
- GPU count
- 72
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
Deployment reality
Strengths, limits and runtime route
What it is good at
- Designed for real-time ASR rather than only batch transcription.
- Apache 2.0 licence.
- Official vLLM library integration.
- Fits within the entry 32GB product's memory envelope.
Where to be cautious
- Real-time operation is sensitive to chunking, network and audio buffering.
- A model-memory fit does not establish supported simultaneous streams.
- Accuracy needs testing by language, accent, noise and domain.
- The runtime and audio API are newer than Whisper's mature ecosystem.
Serving software
- vLLM
- Mistral reference tooling
Before installation
- Measure end-to-end latency, not only model execution.
- Record audio chunk size, sample rate, concurrent streams and dropped or delayed segments.
- Compare against Whisper or Qwen3-ASR on the same labelled audio.
- Pin vLLM and its audio dependencies.
System requirements
- GPU layout
- The minimum can fit on one GPU; multiple GPUs may still be useful for replicas or throughput.
- Context and cache
- The model context ceiling is not a guaranteed serving target. KV cache, batch size and concurrent sessions need separate capacity tests.
- System RAM
- Size system memory for model loading, runtime overhead, preprocessing and any CPU offload used by the final configuration.
- Storage
- Allow space for the pinned checkpoint, runtime images, caches, logs and at least one rollback version.
- Serving software
- Validate the exact checkpoint with vLLM, Mistral reference tooling before acceptance.
- Representative workload
- Benchmark representative prompts or media at the required context, quality, latency and concurrency.
Performance: speed, usable context and concurrency depend on the selected system, software and workload. No benchmark is quoted on this page.
Commercial and legal boundary
Apache License 2.0
Commercial use: permitted
Apache 2.0 permits commercial use subject to notices and lawful audio processing.
Always retain the applicable notices and recheck the live terms for the intended organisation, territory, use and distribution route. Obtain legal advice where required; the official licence governs use.
Read the official licenceTechnical questions
Voxtral Mini 4B Realtime deployment FAQ
Can Team 32 run Voxtral Realtime?
Its memory requirement fits. The required stream count and latency still need testing.
Is Voxtral Realtime a text-to-speech model?
No. This checkpoint is for automatic speech recognition. Voxtral TTS is a separate model family.
Why choose it over Whisper?
Its main attraction is a current real-time streaming design. Whisper may still win for ecosystem maturity or a specific language and noise profile.
What a complete Voxtral Mini 4B Realtime deployment needs
GPU memory is only one part of the system. Storage, data access, serving software, monitoring and administrator handover also affect a reliable deployment.
