# Run Codestral 22B

Built specifically for filling in code in the middle of a file. Ready in about 90 seconds on an RTX 4090, from $0.45/hr. No install, no code.

## What it does

Codestral was trained for a task most coding models handle badly: writing the missing piece when there is code both above and below the gap.

That is what editor completion actually is, and it is why Codestral often feels better in an editor than a larger general model.

The license is restrictive. It is free for personal use, research, and testing, but not for commercial products or internal business use without a paid license from Mistral.

## Your setup

We run Codestral 22B on an RTX 4090 with 24GB of memory, at $0.45 per hour, billed by the minute. It takes about 2 minutes per thousand lines.

At 22 billion parameters the 4-bit version is about 13GB, leaving around 10GB on a 24GB card for context. Fill-in-the-middle work needs less context than a chat assistant, because the useful signal is the code immediately around the gap rather than the whole repository, so this is a comfortable fit rather than a tight one.

## What it costs

30 thousand lines written takes about 68 minutes of GPU time, which is about $0.51.

## Two ways to run it

- **Open the app.** Connect your editor and use it for completion.
- **Run a batch job.** Not the usual mode for this model. Use Qwen2.5-Coder for repository-wide jobs.

## Model facts

| Field | Value |
|---|---|
| Parameters | 22B |
| License | Mistral AI Non-Production License |
| Memory required | 13GB minimum, 24GB recommended |
| Base model | Trained from scratch by Mistral AI |
| Publisher | Mistral AI |
| Recommended hardware | RTX 4090, 24GB |
| Category | [Coding assistants](https://gpuvault.io/working-on/coding-assistants/) |

## Questions people ask

### Can I use Codestral at work?

Not without a paid license. The Mistral AI Non-Production License covers personal projects, research, and evaluation, but explicitly excludes commercial and internal business use. If you are writing code for an employer or a client, use Qwen2.5-Coder 32B or DeepSeek-Coder V2 Lite instead. Both permit commercial use.

### Do I need to install anything to use Codestral 22B?

No. We start a machine with Codestral 22B already loaded and hand you a link. Everything runs in your browser, there is nothing to download, and nothing is left on your computer afterwards. It works the same on a Mac, a Windows laptop, or a Chromebook. A workspace is usually ready in about 90 seconds.

### How much does it cost to run 30 thousand lines written?

About $0.51. Codestral 22B takes about 2 minutes per thousand lines on the RTX 4090 we recommend, so 30 thousand lines written is roughly 68 minutes of GPU time at $0.45 per hour. Billing is by the minute, and a batch job shuts the GPU off the moment the last item finishes. The calculator above works this out for your own numbers.

### Can I run Codestral 22B on a cheaper card?

Sometimes, and the calculator will not always make it look worth it. At 22 billion parameters the 4-bit version is about 13GB, leaving around 10GB on a 24GB card for context. Fill-in-the-middle work needs less context than a chat assistant, because the useful signal is the code immediately around the gap rather than the whole repository, so this is a comfortable fit rather than a tight one. If you want to try a different card anyway, the advanced catalog lets you pick one and shows the estimated time before you commit.

### What are the license restrictions on Codestral 22B?

Codestral 22B is released under the Mistral AI Non-Production License. That is not a permissive license, so read it before you use the output in paid work. We show the license on every model page precisely because this catches people out, and there is usually a permissively licensed alternative in the same category.

### What happens to my files after I stop?

Anything you save into your workspace folder stays in your account and is there when you come back. Everything else is destroyed: the machine is terminated and its disk is wiped. Batch job inputs and outputs are deleted from our storage 24 hours after you download them. We do not read your files or use them to train anything.

---

*Source: https://gpuvault.io/model/codestral-22b/ — GPUVault, GPU rental by the minute.*