Codestral22B params33k contextpopular
Codestral 22B locally
Codestral 22B is an open-weight LLM from the Codestral family with 22B parameters. Main use: code and developer agents. Detected minimum hardware: RTX 4060 Ti 16GB (16 GB).
Technical facts
Parameters22B
Max context33k
Q4_K_M13.8 GB
Q5_K_M16.9 GB
Q824.6 GB
FP1649.2 GB
FamilyCodestral
Last sync2026-05-12
Available quantizations
GGUF weightsQ4_K_M
13.8GB
Acceptable. Good compromise when VRAM is limited.
Q5_K_M
16.9GB
Good quality. Sweet spot for size and precision.
Q8
24.6GB
Near-FP16 quality. Comfortable for production.
FP16
49.2GB
Reference precision. Maximum quality, doubled VRAM.
Compatible GPUs
12 single-GPUGPUs that can run Codestral 22B on a single card, ranked by VRAM headroom.
RTX 4060 Ti 16GB
NVIDIA16 GB · RTX 40
13.8 / 16 GBtight · Q4
RTX 4070 Ti Super
NVIDIA16 GB · RTX 40
13.8 / 16 GBtight · Q4
RTX 4080
NVIDIA16 GB · RTX 40
13.8 / 16 GBtight · Q4
RTX 4080 Super
NVIDIA16 GB · RTX 40
13.8 / 16 GBtight · Q4
RTX 5060 Ti 16GB
NVIDIA16 GB · RTX 50
13.8 / 16 GBtight · Q4
RTX 5070 Ti
NVIDIA16 GB · RTX 50
13.8 / 16 GBtight · Q4
RTX 5080
NVIDIA16 GB · RTX 50
13.8 / 16 GBtight · Q4
Quadro RTX 5000
NVIDIA16 GB · Quadro RTX
13.8 / 16 GBtight · Q4
RTX A4000
NVIDIA16 GB · RTX A (Ampere)
13.8 / 16 GBtight · Q4
RTX 2000 Ada
NVIDIA16 GB · RTX Ada
13.8 / 16 GBtight · Q4
Radeon RX 6800
AMD16 GB · RDNA 2
13.8 / 16 GBtight · Q4
Radeon RX 6800 XT
AMD16 GB · RDNA 2
13.8 / 16 GBtight · Q4
Recommended multi-GPU rigs
2x / 4x consumer GPUsFor Codestral 22B at higher quantization or with more context, a multi-GPU rig gives more headroom.
2× GTX 1070
NVIDIA16 GB · GTX 10
13.8 / 16 GBtight · Q4
2× GTX 1070 Ti
NVIDIA16 GB · GTX 10
13.8 / 16 GBtight · Q4
2× GTX 1080
NVIDIA16 GB · GTX 10
13.8 / 16 GBtight · Q4
4× GTX 1650
NVIDIA16 GB · GTX 16
13.8 / 16 GBtight · Q4
2× RTX 2060 Super
NVIDIA16 GB · RTX 20
13.8 / 16 GBtight · Q4
2× RTX 2070
NVIDIA16 GB · RTX 20
13.8 / 16 GBtight · Q4
2× RTX 2070 Super
NVIDIA16 GB · RTX 20
13.8 / 16 GBtight · Q4
2× RTX 2080
NVIDIA16 GB · RTX 20
13.8 / 16 GBtight · Q4
Recommended rig
2× GTX 1070
Codestral 22B with Ubuntu, vLLM, Open WebUI and the model already downloaded.
VRAM estimates: parameters x bits/8 plus margin. Real performance varies by engine, context length and batch size.
sync: 2026-05-12