Skip to content

Devstral models

Mistral's coding models: Devstral agents that explore codebases and edit many files, and Codestral for autocomplete.

  • Open weights
  • Models 4
  • Context up to 384K
  • Updated Nov 2026
Hugging Face Compare
On this page

At a glance

Type
Open-weight models
Developer
Mistral AI (France)
First release
May 2024
Models tracked
4 (2 current)
Longest context
384K tokens
Abilities
Tool callingImage input
Licence
Devstral Small models are Apache 2.0. Devstral 2 (123B) uses a modified MIT licence. Codestral 22B uses the Mistral Non-Production License (no commercial use).
Smallest local option
Devstral Small 1 24B · ~16 GB
Website
mistral.ai
Hugging Face
huggingface.co/mistralai

Newest model

Devstral 2

Released Dec 2025 · 256K context

Smallest option

Devstral Small 1 24B

~16 GB VRAM · 14 GB download

Devstral models and versions

ModelReleasedSizesContextAbilitiesLicenceStatus
Devstral 2 Large agentic coding model, 256K contextDec 2025 123B256KToolsModified MIT Current
Devstral Small 2 Coding agent on one GPU, image input, 384K contextDec 2025 24B384KToolsVisionApache 2.0 Current
Devstral Small 1 First Devstral, built with All Hands AIMay 2025 24B128KToolsApache 2.0 Previous generation
Codestral 22B Autocomplete (fill-in-the-middle); non-commercial licenceMay 2024 V0.1 · 22B32KText onlyMistral Non-Production License Legacy

Sizes, context windows and abilities come from the Ollama library and Hugging Face model cards, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

How much VRAM does Devstral need?

Download sizes per Ollama build and the GPU memory each needs with an 8K context. On a CPU-only server the same number is the RAM you need. Longer contexts need more memory; work it out with the LLM VRAM calculator.

Model4-bit (default)8-bit16-bitVRAM (4-bit)Fits onCommand
Devstral 2 123B75 GB133 GB250 GB~83 GB128 GB GPU
Devstral Small 2 24B15 GB26 GB48 GB~17 GB24 GB GPU
Devstral Small 1 24B14 GB25 GB47 GB~16 GB16 GB GPU

VRAM = 4-bit download + about 10% runtime overhead + 0.5 GB for an 8K context, rounded up. Mixture-of-experts (MoE) models still need all their weights in memory, but they run much faster than a dense model of the same size, so they also work well split between GPU and system RAM.

Run Devstral 2 locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run devstral-2:123b

New to Ollama? Follow Install Ollama on Ubuntu 26.04 (or 24.04) for GPU drivers, the systemd service and remote access.

What Devstral is good for

Coding agents

Devstral Small 2 works well in OpenHands, Cline and Mistral Vibe on a 16–24 GB GPU.

Large refactors on a server

Devstral 2 (123B) handles multi-file changes on a multi-GPU box.

Screenshots to code

Devstral Small 2 takes images, so it can read UI mock-ups and error screenshots.

Strengths and limits

Strengths

  • Trained for agent tools: searching code, editing files, running tests.
  • Devstral Small 2 is Apache 2.0 with a 384K-token context.

Watch out for

  • Mistral now uses Medium 3.5 instead of Devstral 2 in its own Vibe agent.
  • Codestral 22B cannot be used commercially.

About Devstral

Devstral is Mistral AI’s family of models for software-engineering agents. The first Devstral (May 2025), built with All Hands AI, was a 24B model trained to use the tools a coding agent needs: search the codebase, open and edit files, run commands and react to test failures.

The second generation (December 2025) comes in two sizes: Devstral Small 2, a 24B Apache 2.0 model with image input and a 384K context, and Devstral 2, a 123B model under a modified MIT licence. In 2026 Mistral replaced Devstral 2 with Mistral Medium 3.5 inside its own Vibe coding agent, but the Devstral weights remain available.

Codestral (May 2024) is Mistral’s older autocomplete model; its weights use a non-production licence, so for commercial autocomplete use Qwen2.5-Coder instead.

Guides for running Devstral

Tool LLM VRAM Calculator: How Much GPU Memory Do You Need? Updated Nov 2026 AI & LLMs How to Install Ollama on Ubuntu 26.04 and Run LLMs Locally Updated Nov 2026 AI & LLMs How to Install llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) Updated Nov 2026 AI & LLMs How to Install vLLM on Ubuntu 26.04 (OpenAI-Compatible LLM Server) Updated Nov 2026 AI & LLMs How to Install Open WebUI on Ubuntu 26.04 (Self-Hosted ChatGPT Alternative) Updated Nov 2026

Frequently asked questions

What is the latest Devstral model?

The newest Devstral model is Devstral 2, released in Dec 2025. It has a 256K-token context window and supports tool calling.

How much VRAM do I need to run Devstral locally?

The smallest option, Devstral Small 1 24B, is a 14 GB download at 4-bit and needs about 16 GB of GPU memory (or RAM on a CPU-only machine) with an 8K context. Larger sizes need more; see the hardware table on this page, and use the VRAM calculator for longer contexts.

How do I run Devstral with Ollama?

Install Ollama, then run "ollama run devstral:24b". Ollama downloads the 4-bit build and starts a chat; the same model is also available on the local API at http://localhost:11434.

Can I use Devstral commercially?

Devstral Small models are Apache 2.0. Devstral 2 (123B) uses a modified MIT licence. Codestral 22B uses the Mistral Non-Production License (no commercial use). Always read the licence of the exact model you deploy; it is linked in the models table.

Compare Devstral

← All AI models

Back to top