Codestral 2 (22B)

Mistral Apr 2026; next-gen code specialist, FIM support, beats GPT-4o on HumanEval and MBPP. First Codestral under Apache 2.0 — previous versions were non-commercial MNPL.

Model Specifications

ArchitectureTEXT

Parameters22B

Familycodestral

VRAM (Q4)11.0GB

Apache 2.0 licensed. 256K context. 380K HF downloads in first week. Runs on single RTX 4090 at Q4 (~13GB VRAM).

#mistral#coding#fim#apache2#open-source#ideSource

Estimated Quantization Sizes

Format	Precision	Est. VRAM	Recommendation
FP16 / BF16	16-bit	44.0 GB	Uncompressed Base
Q8_0High	8-bit	22.0 GB	Near Lossless
Q6_K	6-bit	16.5 GB	Excellent Balance
Q4_K_MPopular	4-bit	11.0 GB	Standard Use

Share this Model

Send this model's specs directly to your community.

Post

Similar Models

Llama 3.3 70B

70.55B

Refined Llama 3 with superior following

Llama 3.2 3B

3.21B

Mobile-optimized small model

Llama 3.2 1B

1.23B

Ultra-light edge deployment

Related Guides

How much VRAM do you really need?

A complete breakdown of quantization levels and VRAM overhead for running local models.

Best GPUs for Machine Learning in 2026

Comparing NVIDIA and AMD options for the best speed-to-dollar ratio.

GGUF vs EXL2 vs AWQ

Understanding local AI formats and which one to pick for your specific hardware.