The Open-Source AI Stack
RSS
All models

Models · mistral

Codestral 22B v0.1

Source-available Mistral AI · 2024-05-29 · Mistral AI Non-Production License (MNPL-0.1)

Mistral's first code-specialist release, weights-available in May 2024 under the Mistral AI Non-Production License (research and testing only, commercial requires a separate license). Trained on 80+ programming languages with a 32K context, four to eight times longer than contemporary code models, and supports both instruct-style answering and fill-in-the-middle completion for IDE integrations.

Cost

— / Mtok input
— / Mtok output

Mistral La Plateforme · as of 2026-05-19

via Artificial Analysis ↗

Architecture

tokens in Embedding vocab not disclosed · mistral tokenizer × N layers Grouped-Query Attention RoPE context 32,768 tokens Dense MLP SwiGLU activation (standard) 22B active params Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
dense
Total params
22B
Active params
22B
Context window
33K tokens
Attention
gqa
Position encoding
rope
Post-training
sft
OSI-approved
no
Data released
no
Training code
not released

Available quantizations

GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8. runs on llama.cpp, Ollama
EXL2 ExLlamaV2's variable-bitrate format for consumer GPUs. runs on ExLlamaV2
MLX Apple MLX 4/8-bit layout for Apple silicon. runs on Apple MLX

Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.

Notable innovations

  • · 32K context for repo-scale code completion
  • · Fill-in-the-middle endpoint for IDE plugin use cases
  • · Coverage of 80+ programming languages

Sources