Evidence: benchmark results vendor-reported
Shieldstral 1.0 3B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
A 3.8B safety classifier on the Ministral 3 layout. Each request carries your policy as a question; the model answers with a safety score you can threshold or rank. Trained on sequences up to 32K tokens. Apache-2.0; Mistral says it runs on a single 16 GB GPU in BF16.
Shieldstral 1.0 3B needs about 3 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 2.9 GB | ~222 tok/s | Fits | |
| Q3_K_M 3.41 bpw | 3.3 GB | ~200 tok/s | Fits | |
| Q4_K_M 4.83 bpw | 4 GB | ~170 tok/s | Fits | |
| Q5_K_M 5.67 bpw | 4.4 GB | ~155 tok/s | Fits | |
| Q6_K 6.56 bpw | 4.8 GB | ~143 tok/s | Fits | |
| Q8_0 8.50 bpw | 5.8 GB | ~121 tok/s | Fits | |
| F16 16.00 bpw | 9.4 GB | ~77 tok/s | Fits |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Shieldstral 1.0 3B VRAM calculator →
Get Shieldstral 1.0 3B running
The cheapest catalogued GPU that runs Shieldstral 1.0 3B is the Intel Arc B570 (10 GB).
or compare on Vast.ai from $0.35/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Shieldstral 1.0 3B
No Ollama tag. Mistral documents vLLM and llama.cpp builds on the model card.
Specifications
Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.
- Parameters
- 3.8 Billion
- Context window
- 32,768
- Architecture
- Dense Transformer (GQA), image input
- Provider
- Mistral AI
- Licence
- Apache-2.0
- Practical deployment
- Edge device (phone, SBC)
- Specified at
- Q4_K_M
- System RAM
- 16 GB
- Record updated
- 2026-11-03
Commercial use permitted. No usage restrictions beyond attribution.
Limits and caveats
- A classifier, not a chat model: it scores content against the policy you pass in.
- Trained on sequences up to 32K tokens; the card advises staying inside that range.
See also Ministral 3 3B
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Shieldstral 1.0 3B — frequently asked questions
How much VRAM does Shieldstral 1.0 3B need?
About 3 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Shieldstral 1.0 3B run on an RTX 4090 (24 GB)?
Yes. Shieldstral 1.0 3B needs about 3 GB at Q4_K_M, inside a 24 GB card, at an estimated 170 tokens/sec.
How do I run Shieldstral 1.0 3B locally?
No Ollama tag. Mistral documents vLLM and llama.cpp builds on the model card. Running the published tag would send your prompts to a hosted GPU rather than your own machine.