Codestral 22B

mistralai/Codestral-22B-v0.1

Parameters

22B

Context

32K

Downloads

39.3K

Likes

1.3K

Architecture

transformer

View on HuggingFace

GPU Requirements

VRAM / RAM / disk estimates by quantization and usage scenario

QuantizationMinimumRecommendedProduction
FP1616bit
56.8 GB
RAM 85.2 · Disk 102.9
74.9 GB
RAM 112.3 · Disk 128.9
174.2 GB
RAM 261.2 · Disk 167.9
FP88bit
30.9 GB
RAM 46.3 · Disk 71.0
47.8 GB
RAM 71.7 · Disk 97.0
145.9 GB
RAM 218.8 · Disk 136.0
INT88bit
30.9 GB
RAM 46.3 · Disk 71.0
47.8 GB
RAM 71.7 · Disk 97.0
145.9 GB
RAM 218.8 · Disk 136.0
AWQ4bit
18.4 GB
RAM 32.0 · Disk 55.6
34.8 GB
RAM 52.1 · Disk 81.6
132.3 GB
RAM 198.4 · Disk 120.6
GPTQ4bit
18.6 GB
RAM 32.0 · Disk 55.8
34.9 GB
RAM 52.4 · Disk 81.8
132.4 GB
RAM 198.7 · Disk 120.8
GGUF4bit
18.9 GB
RAM 32.0 · Disk 56.2
35.3 GB
RAM 52.9 · Disk 82.2
132.8 GB
RAM 199.2 · Disk 121.2

GPU Recommendations

Best ValueBEST

1× H100 SXM

80 GB VRAM

$1,170/mo

≈ 315.1 tok/s

Effective production throughput ≈ 220.6 tok/s

score 0.6054

Deploy on vast ↗
Cheapest

4× Tesla V100

64 GB VRAM

$84.4/mo

≈ 0 tok/s (N/A — benchmark unavailable)

Effective production throughput ≈ 0 tok/s

score 0.5431

Deploy on vast ↗
Performance

8× H200 SXM

1128 GB VRAM

$20,440/mo

≈ 1445 tok/s

Effective production throughput ≈ 1011.5 tok/s

score 0.2197

Min Complexity

1× RTX A6000

48 GB VRAM

$210/mo

≈ 72.2 tok/s

Effective production throughput ≈ 50.5 tok/s

score 0.6842

Deploy on vast ↗

💬 Community — real deployment experience

0 Articles · 0 Benchmarks

View Community →

GPUs that fit (single card)

RTX A6000

48 GB VRAM

$210

/mo · vast

L40S

48 GB VRAM

$341

/mo · vast

RTX PRO 6000 S

48 GB VRAM

$633

/mo · vast

RTX PRO 6000 WS

48 GB VRAM

$634

/mo · vast

A100 80GB

80 GB VRAM

$876

/mo · vast

AmciHub

AI Model Deployment & Compute Intelligence — analyze AI model requirements, GPU performance, cloud pricing and deployment costs to find the right deployment solution.

© 2026 AmciHub. All rights reserved. Model → Requirement → GPU → Benchmark → Cloud → Cost → Recommendation