# Devstral Small 2 (slim)

> Devstral Small 2 is Mistral's open-weight coding agent model, 24B parameters with image input, tool calling and a 256K context. Apache 2.0. It runs in vLLM, SGLang, Transformers, llama.cpp and Ollama, where the 4-bit build is 15 GB.

- Full: https://www.anchorterminal.com/tools/devstral-small-2.md (~10,050 tokens) · this version ~1,630 tokens · JSON https://www.anchorterminal.com/tools/devstral-small-2.json · canonical https://www.anchorterminal.com/tools/devstral-small-2
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-10 · pricing https://www.anchorterminal.com/tools/devstral-small-2#pricing · reviews https://www.anchorterminal.com/tools/devstral-small-2#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #60 of 160 open-weight models · #50 in Open-weight models · not agent-ready · confidence low** · same model listed separately: Mistral AI API, Model API (https://www.anchorterminal.com/tools/mistral-api.min.md)


Assessment: The easy one to run: a 24B coding agent under Apache 2.0 that five runtimes carry, 15 GB at 4 bits in Ollama, with a 256K context, image input and well-documented tool calling. It's eleven months old, calls tools one at a time, has no reasoning mode, and its card says nothing on limits, safety or data.

## Facts

- Kind: Open-weight model · vendor: Mistral AI · category: Open-weight models · legal entity: Mistral AI (RCS Paris 952 418 325) · provenance 84/100
- Packages: huggingface `mistralai/Devstral-Small-2-24B-Instruct-2512`, ollama `devstral-small-2`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0
- Benchmark probes: not run yet, so Performance is pending
- Surface graded: The open weights at huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512, read from the model card, the files and their metadata, the config, the chat template, the commit history, ggml-org's GGUF repository, Ollama's library tags and the runtimes' own code at their latest releases on 10 November 2026. Mistral's company facts come from our Mistral Large 4 listing. The two system prompt files, the test prompts and LM Studio's catalogue page weren't read. Nothing was downloaded or run
- Model: 24,011,361,280 parameters in the files. A dense 40-layer model on the architecture of Ministral 3, with a Pixtral vision encoder for image input, fine-tuned from Mistral-Small-3.1-24B-Base for agentic coding. Mistral3ForConditionalGeneration
- Files: Safetensors in FP8, twice over: a consolidated set for mistral-common and vLLM and a model set for Transformers, 25.79 GB each. chat_template.jinja, config, tokenizer and tekken files, params.json and two system prompt files. No pickle files and no Python. Hugging Face lists a SHA-256 for the large files
- Licence: Apache 2.0, in the card's metadata and its text, which says commercial and non-commercial use are both allowed
- Runtimes: The card documents vLLM (recommended), SGLang and Transformers, and links llama.cpp, LM Studio and Ollama. Released vLLM 0.33.0, SGLang 0.5.23, Transformers 5.22.2 and llama.cpp b12242 carry the mistral3 architecture, ggml-org publishes a GGUF, and Ollama's library has devstral-small-2
- Context: 256K tokens per the card, and its vLLM command serves 262,144. config.json gives 393,216 positions, and Ollama's library lists 384K
- Tools and reasoning: Tools in the OpenAI function format, with the mistral tool-call parser and auto tool choice on vLLM. The card's test prompts check one tool call and tool calls made one after another. No reasoning mode: a June commit removed stray thinking tokens
- Hardware: A single RTX 4090 or a Mac with 32 GB of memory, per the card. Its vLLM command uses two GPUs for the full context
- Recommended settings: generation_config.json sets temperature 0.15. The card says to use the system prompt in CHAT_SYSTEM_PROMPT.txt, or VIBE_SYSTEM_PROMPT.txt with Mistral's Vibe agent
- Training data: Not described on the card

- Scores: Capability pending, Reach 90, Risk 60, Agent interface 65, Documentation and upkeep 46 · total over the 7 assessed categories
- Why:  · Reach, Memory 25, the smallest build at 4 bits from a runtime's own library is Ollama's devstral-small-2 at Q4_K_M, 15 GB, and 15 GB × 1.2 = 18 GB… · Risk, File safety 25, safetensors only, no .bin, .pt or .pth files and no Python in the repository. · Agent interface, Chat template 20, chat_template.jinja in the repository (11,362 characters), with the tokenizer's tekken file for mistral-common. · Documentation and upkeep, Model card 11, Intended use, the architecture, the parameters and the context are on the card.
- Sources: 14, open questions: 6, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/devstral-small-2.json
- Its vendor verifies the listing with `verify_listing` or `POST https://www.anchorterminal.com/api/v1/verify` (the full twin says how); no effect on the grade.
- Support: Discord https://discord.gg/mistralai; https://help.mistral.ai/en/articles/347458-how-do-i-contact-support; https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/discussions; more in the full twin

## Before you call it

1. `ollama run devstral-small-2` is the quickest start, at 15 GB
2. On vLLM pass `--tool-call-parser mistral --enable-auto-tool-choice`
3. Send the system prompt from CHAT_SYSTEM_PROMPT.txt in the repository. The card says results depend on it
4. Use a low temperature. The repository's default is 0.15
5. Plan for one tool call per turn
6. The same model is on our Mistral API listing if you'd rather not host it

## Connect

```bash
ollama run devstral-small-2
# or: vllm serve mistralai/Devstral-Small-2-24B-Instruct-2512 --max-model-len 262144 --tool-call-parser mistral --enable-auto-tool-choice
```

```bash
# vLLM serves an OpenAI-compatible API at http://localhost:8000/v1, and Ollama at http://localhost:11434/v1.
# Send the system prompt from CHAT_SYSTEM_PROMPT.txt and tools in the OpenAI function format
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 75.4 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 73.6 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 12B IT | BB | 72.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)