Mistral AI

Codestral 22B

Mistral AI·22B·Mistral·32K ctx·2026-08

Check on my hardwareAdd to compare →Ollama →Hugging Face →

Model size→Hardware requirement→Actual fit

Codestral 22B needs ~14.2–26 GB of VRAM depending on quantization. The bar is what it needs; the marker is the card's VRAM. Speed figures below are planning estimates.

8 GB

RTX 4060 · 3060 Ti

14.2
8
Q4_K_M · Est. ~3 tok/sSlow · CPU offload

12 GB

RTX 4070 · 3060 12G

14.2
12
Q4_K_M · Est. ~5 tok/sSlow · CPU offload

16 GB

RTX 4080 · M-series 16G

14.2
16
Q4_K_M · Est. ~55 tok/sComfortable

24 GB

RTX 4090 · 3090

16.6
24
Q5_K_M · Est. ~69 tok/sComfortable

48 GB

A6000 · 2×24G

26
48
Q8_0 · Est. ~40 tok/sComfortable

02  — Quantizations

QuantFile sizeEst. VRAM
Q4_K_M13 GB14.2 GB
Q5_K_M15.4 GB16.6 GB
Q8_024.8 GB26 GB

03  — Runtimes & use

llama.cpp · Ollama · LM Studio

Coding

Memory and speed figures are planning estimates, not measured benchmarks. Memory includes a fixed allowance for KV cache and runtime overhead. Long-context prompts push VRAM use above the numbers shown here.