Ministral 3 8B VRAM Requirements

Ministral 3 8B needs about 6.7 GB of VRAM at Q4_K_M with an 8K context: 4.8 GB for weights, 1.1 GB for the KV cache and 0.8 GB of runtime overhead. At Q8_0 it needs 10.3 GB and at full FP16 precision 17.7 GB. The smallest GPU that runs it fully is the RTX 3050 8GB.

By Mistral AI · Updated October 14, 2026 · Values from config.json on Hugging Face

6.7 GB VRAM

Ministral 3 8B at Q4_K_M with 8K context and the KV cache in FP16.

  • Weights 4.8 GB
  • KV cache 1.1 GB
  • Runtime overhead 0.8 GB

Which GPUs can run it

Ministral 3 8B VRAM by quantization

Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.

QuantizationBits/weightWeightsTotal at 8KTotal at 32KSmallest tier
Q4_K_M4.854.8 GB6.7 GB9.9 GB8 GB
Q5_K_M5.695.6 GB7.5 GB10.7 GB8 GB
Q6_K6.596.5 GB8.4 GB11.6 GB10 GB
Q8_08.508.4 GB10.3 GB13.5 GB12 GB
FP16 / BF1616.0015.8 GB17.7 GB20.9 GB20 GB
Q3_K_M3.913.9 GB5.7 GB8.9 GB8 GB
Q2_K3.003.0 GB4.8 GB8.0 GB6 GB

VRAM by context length

At Q4_K_M. The KV cache grows linearly with context: 0.13 GB per 1,000 tokens in FP16.

ContextKV cache (FP16)KV cache (Q8_0)Total, FP16 cacheTotal, Q8_0 cache
4K tokens0.5 GB0.3 GB6.1 GB5.9 GB
8K tokens1.1 GB0.6 GB6.7 GB6.2 GB
32K tokens4.3 GB2.3 GB9.9 GB7.9 GB
128K tokens17.0 GB9.0 GB22.6 GB14.6 GB
256K tokens34.0 GB18.1 GB39.6 GB23.7 GB

GPUs that can run Ministral 3 8B

At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.

GPUUsable memoryStatusBest quantMax contextSpeed
RTX 3050 8GB8.0 GBFitsQ5_K_M14K~25 tok/s
RTX 4060 8GB8.0 GBFitsQ5_K_M14K~31 tok/s
RTX 4060 Ti 8GB8.0 GBFitsQ5_K_M14K~32 tok/s
Radeon RX 7600 8GB8.0 GBFitsQ5_K_M14K~27 tok/s
RTX 3060 Ti 8GB8.0 GBFitsQ5_K_M14K~50 tok/s
RTX 3070 8GB8.0 GBFitsQ5_K_M14K~50 tok/s
RTX 5060 8GB8.0 GBFitsQ5_K_M14K~50 tok/s
RTX 5060 Ti 8GB8.0 GBFitsQ5_K_M14K~50 tok/s
RTX 3070 Ti 8GB8.0 GBFitsQ5_K_M14K~68 tok/s
Intel Arc B570 10GB10.0 GBFitsQ6_K28K~33 tok/s
RTX 3080 10GB10.0 GBFitsQ6_K28K~84 tok/s
Apple M4 16GB10.7 GBFitsQ6_K33K~14 tok/s
RTX 2080 Ti 11GB11.0 GBFitsQ8_035K~68 tok/s
RTX 2060 12GB12.0 GBFitsQ8_042K~38 tok/s

About Ministral 3 8B

A dense 8B model with vision and a 256K context, released alongside Mistral Large 3. It is a drop-in alternative to Llama 3.1 8B with a similar memory profile: 136 KB of cache per token in FP16. Base, instruct and reasoning variants share the same architecture.

SpecValue
Parameters8.49B
Layers34
Hidden size4096
Attention heads / KV heads32 / 8
Head dimension128
Vocabulary131,072
Max context (used here)256K tokens
KV cache per token (FP16)136 KB
Released2025-12
LicenseApache 2.0

Ministral 3 8B VRAM: common questions

Can Ministral 3 8B run on an 8GB GPU?

Yes at Q4_K_M with up to about 14K of context, similar to Llama 3.1 8B. A Q8_0 cache roughly doubles that.

How much VRAM does Ministral 3 8B need?

About 6.7 GB at Q4_K_M with an 8K context, 10.3 GB at Q8_0 and 17.7 GB at FP16. Longer contexts add 0.13 GB per 1,000 tokens with an FP16 cache.

What is the cheapest GPU that can run Ministral 3 8B?

The smallest listed device that fits it fully at Q4_K_M with 8K context is the RTX 3050 8GB (8.0 GB usable).

Related