Model · Mistral AI · Code
Devstral Small 2 (24B)
Mistral's small open-weight coding-agent model — 24B parameters under Apache 2.0, 256K context, light enough for a single RTX 4090 or a 32 GB Mac. Built to explore codebases, edit many files and drive coding agents.
- Modality
- Code
- License
- Apache 2.0 (Open)
- Parameter size
- 24B (dense)
- Context window
- 256,000 tokens 256k per Mistral's model card.
- Released
- December 9, 2025
- Last verified
- October 28, 2026
- Runs locally
- Yes
- Also handles
- text, vision
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 32 GB (48 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 24 GB (32 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Mistral says it runs on a Mac with 32 GB of memory or a single 24 GB graphics card, and 4-bit builds are available for Ollama, LM Studio and MLX.
Numbers are for: 4-bit build of the 24B (about 15 GB). Based on the maker’s own figures; where the maker gives only a graphics-card figure, the Mac memory is our estimate.
Strengths
- Apache 2.0 — commercial and non-commercial use allowed
- Runs on a single RTX 4090 (24 GB) or a Mac with 32 GB of memory, per Mistral
- 68.0% on SWE-bench Verified (Mistral's own figure), close to models many times its size
- Image input as well as text, new over Devstral Small 1.1
- Works with Mistral Vibe CLI, Cline, Kilo Code, Claude Code, OpenHands and SWE-agent
Weaknesses
- Mistral retired its own API for it (labs-devstral-small-2512) on 2026-03-31 and names Ministral 3 8B as the hosted alternative — today it is a self-host model only
- The published checkpoint is FP8; smaller GGUF and MLX quantisations come from the community
- A coding specialist, not a general chat model
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | Apache 2.0; community GGUF and MLX builds also available |
| Ollama | local | Quantised builds |
| Mistral Vibe | local | Mistral's terminal coding agent |
Used in solutions
Version history
- Devstral 2 Dec 2025 Deprecated
- Devstral Small 2 (24B) Dec 2025 Current
Official sources
- Model card model
- Announcement announcement
- Mistral changelog docs
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: Mistral model card; Ollama library page for devstral-small-2.
- — Initial entry. Added 2026-10-28 from the popularity sweep: 182k Hugging Face downloads/month for the main checkpoint, plus community GGUF and MLX builds. Released 2025-12-09 alongside Devstral 2, per Mistral's changelog. Apache 2.0, 256k context, free to self-host. Mistral's models page lists the API version labs-devstral-small-2512 as deprecated 2026-02-27 and retired 2026-03-31 (alternative: Ministral 3 8B); the weights stay downloadable and there is no newer Devstral Small, so the entry is kept current as a self-host model.
Esc