Codestral 22B v0.1
By Mistral AI · France
Updated 2026-07-13
Overview
Mistral AI's 22B code specialist covering 80+ programming languages, with strong HumanEval and MBPP scores. Locked behind the restrictive MNPL license — personal and research use only.
When to pick this model
- Personal coding assistant on a workstation with 16–24GB VRAM
- Academic research on code generation and completion
- Internal experimentation before committing to a commercial license
- Polyglot codebases where coverage across 80+ languages matters
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 13 GB |
| Q5_K_M | 16 GB |
| Q8_0 | 24 GB |
| FP16 (no quantization) | 44 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, Codestral 22B v0.1 needs a 16 GB card at Q4_K_M (13 GB). Stepping up to Q8_0 nearly doubles the footprint to 24 GB, and unquantized FP16 weights take 44 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, Codestral 22B v0.1 needs roughly 22 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 4 tokens/sec on entry-level GPUs, on the order of 16 tokens/sec on a mid-range card, and up to 42 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches Codestral 22B v0.1 to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for Codestral 22B v0.1 |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Does not fit — needs 13 GB at Q4_K_M |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Does not fit — needs 13 GB at Q4_K_M |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | Q5_K_M (16 GB used) |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | Q8_0 (24 GB used) |
| 32 GB | RTX 5090 | Q8_0 (24 GB used) |
Which GPU should you buy to run Codestral 22B v0.1?
To run Codestral 22B v0.1 locally at Q4, you need ~13 GB of VRAM. The best value for this is a RTX 5070 Ti (16 GB VRAM).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Published benchmark scores
| Benchmark | Score |
|---|---|
| HumanEval | 81.1 |
| MBPP | 78.2 |
| Spider | 63.5 |
Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.
To put Codestral 22B v0.1 in context: its HumanEval score of 81.1 ranks #13 of the 26 catalog models with a published HumanEval result (catalog median 81.1). Scores are author-published and measured under different harnesses, so treat gaps of a few points as noise rather than a verdict.
Strengths
- HumanEval 81.1 and MBPP 78.2 — competitive with much larger models at release
- Broad language coverage including niche languages
- 32k context handles most repo files comfortably
- Strong fill-in-the-middle completion
Limitations
- MNPL license blocks all production and commercial use
- Outclassed by Qwen 2.5 Coder 14B for permissive-licensed alternatives
- 32k context is tight for large-repo agents
Typical workloads
In our catalog grid, Codestral 22B v0.1 is filed under Code Autocomplete, Refactor — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); French-language output where quality matters.
Note the 31k-token context window — fine for short interactions, limiting for long documents or big retrieval contexts. It ships under the Mistral Non-Production License license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: Dense 22B · code-specialized · 80+ languages
Training: Multilingual code corpus.
Capable code model held back by its non-production license — for anything you'd ship, pick Qwen 2.5 Coder 14B instead.
Quick start
ollama run codestral:22bOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does Codestral 22B v0.1 need?
At the recommended Q4_K_M quantization, Codestral 22B v0.1 needs about 13 GB of VRAM. Q8_0 takes 24 GB, and unquantized FP16 weights take 44 GB.
Can Codestral 22B v0.1 run without a GPU?
Yes — with roughly 22 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does Codestral 22B v0.1 support?
Codestral 22B v0.1 supports a 31k-token context window (32,000 tokens).
Can I use Codestral 22B v0.1 commercially?
Codestral 22B v0.1 ships under the Mistral Non-Production License license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is Codestral 22B v0.1 on consumer hardware?
Our compatibility engine estimates on the order of 16 tokens/sec on a mid-range GPU and up to 42 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of Codestral 22B v0.1 should I download first?
Start with Q4_K_M (13 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. On a 24 GB card you can run up to Q8_0.