Best larger-system route
Frontier Native 2.3TB
The smallest system in the range that reaches this model's preferred working allowance.
Explore Frontier Native 2.3TBMistral AI model
Mistral Small 4 Hardware Requirements & Compatibility
The official release contains about 241.9GB of weights. Treat 256GB aggregate as a tight sharded floor and 384GB as the first practical product envelope in this range.
- Model version
mistralai/Mistral-Small-4-119B-2603- Family and variant
- Mistral Small 4 · Mistral-Small-4-119B-2603
- Source version
a11f36bebf70- Source updated
- 15 July 2026
Mistral AI Mistral Small 4
mistralai/Mistral-Small-4-119B-2603 - Minimum GPU memory
- 256GB aggregate with supported model parallelism
- Recommended hardware
- Four 96GB GPUs or a larger coherent platform
- Licence
- Apache License 2.0
- Useful for
- Language & reasoning, Coding & agents, Vision & OCR
Buyer verdict
Where Mistral Small 4 is a sensible fit
Mistral Small 4 suits organisations that want one Apache-licensed model for general chat, visual inputs, reasoning and coding-agent work on an enterprise multi-GPU server.
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
- 256GB aggregate with supported model parallelism
- Recommended
- Four 96GB GPUs or a larger coherent platform
This is close to the repository total and leaves little practical context or runtime headroom.
The recommendation favours operating headroom and fewer model-sharding boundaries.
Technical specification
Mistral Small 4 model and hardware facts
Specifications shown for source version
a11f36bebf709121056b1dbcc943d1c6afbe494d, updated
15 July 2026.
- Architecture
- MoE, 119B total / 6B active
- Context
- 256K tokens
- Modalities
- Text + image → text
- Repository weights
- 241.85GB
- Runtime
- vLLM recommended
- Licence
- Apache 2.0
Product compatibility
Mistral Small 4 compatibility across all 11 GPU systems
- Systems that fit without model splitting
- 2
- Systems needing multi-GPU validation
- 4
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
This system provides 32GB per GPU and 32GB in total. The model requires 256GB aggregate across at least 4 GPUs.
sme workstation
Company 64
- Per GPU
- 32GB
- Total GPU memory
- 64GB
- GPU count
- 2
This system provides 32GB per GPU and 64GB in total. The model requires 256GB aggregate across at least 4 GPUs.
sme workstation
Studio 96
- Per GPU
- 96GB
- Total GPU memory
- 96GB
- GPU count
- 1
This system provides 96GB per GPU and 96GB in total. The model requires 256GB aggregate across at least 4 GPUs.
sme workstation
Studio 192
- Per GPU
- 96GB
- Total GPU memory
- 192GB
- GPU count
- 2
This system provides 96GB per GPU and 192GB in total. The model requires 256GB aggregate across at least 4 GPUs.
pcie rack
Value Rack 128
- Per GPU
- 32GB
- Total GPU memory
- 128GB
- GPU count
- 4
This system provides 32GB per GPU and 128GB in total. The model requires 256GB aggregate across at least 4 GPUs.
pcie rack
Value Rack 256
- Per GPU
- 32GB
- Total GPU memory
- 256GB
- GPU count
- 8
Total installed memory is sufficient, but the model must be divided across GPUs. Confirm that the serving software supports this GPU layout and test the required context and speed.
pcie rack
Enterprise 384
- Per GPU
- 96GB
- Total GPU memory
- 384GB
- GPU count
- 4
Total installed memory is sufficient, but the model must be divided across GPUs. Confirm that the serving software supports this GPU layout and test the required context and speed.
pcie rack
Enterprise 768
- Per GPU
- 96GB
- Total GPU memory
- 768GB
- GPU count
- 8
Total installed memory is sufficient, but the model must be divided across GPUs. Confirm that the serving software supports this GPU layout and test the required context and speed.
pcie rack
H200 1.1TB
- Per GPU
- 141GB
- Total GPU memory
- 1,128GB
- GPU count
- 8
Total installed memory is sufficient, but the model must be divided across GPUs. Confirm that the serving software supports this GPU layout and test the required context and speed.
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
- Unified instruct, reasoning and coding-agent behaviour in one checkpoint.
- Sparse 119B A6B architecture and Apache 2.0 licence.
- Multimodal input and a documented 256K context window.
- Official vLLM deployment and fine-tuning guidance.
Where to be cautious
- The named checkpoint is too large for one 96GB GPU.
- A 256GB total across eight 32GB PCIe GPUs is not equivalent to a coherent 256GB accelerator.
- Vision inputs and long contexts add material memory beyond the base weights.
- Runtime support is fast-moving and should be pinned to the version used during acceptance.
Serving software
- vLLM
- Transformers
- Mistral reference tooling
Before installation
- Prefer four or more 96GB GPUs over many 32GB cards when model sharding and communication dominate.
- Test instruct and reasoning modes separately because output length and latency differ.
- Record image resolution and count for visual workloads.
- Use the model's official prompt and tool formats rather than assuming generic Mistral behaviour.
System requirements
- GPU layout
- Plan for at least 4 GPUs and validate model parallelism on the final topology.
- 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, Transformers, 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 its notice and attribution provisions.
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
Mistral Small 4 deployment FAQ
What is the smallest GPU Servers product for Mistral Small 4?
Value Rack 256 clears the raw aggregate file-size floor, but its eight separate 32GB cards require supported multi-GPU loading. Enterprise 384 is the first more credible planning fit.
Does 6B active mean it fits like a 6B model?
No. Active parameters affect compute per token, but the 119B checkpoint weights still need to be stored and addressed.
Can Mistral Small 4 process images?
Yes, the official model is multimodal. Include image count, resolution and context in the workload test.
What a complete Mistral Small 4 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.
