Mistral Small 3.2
Mistral Small 3.2 comes from Paris-based Mistral AI and is released under the Apache 2.0 licence. It runs on a computer with 32 GB, understands text and images and ranks near the top of EuroEval’s German comparison. For companies that prefer a European maker, it is the natural place to start.
Profile
- Maker
- Mistral AI · France
- Licence
- Free for commercial use
- Runs on
- Computer with 32 GB
- Memory at 4 bits
- 16 GB
- Memory at 8 bits
- 25 GB
- Parameters
- 24bn
- Context
- 33,000 tokens
- Inputs
- text, image
- Released
- June 2025
Scores
- Human rating
- 1338 · rank 88 of 226 open
- In German
- 1340 · rank 73 of 154 open
- EuroEval
- German: rank 4 of 140
- Epoch index
- 131.8 · rank 55 of 127 open
- GPQA Diamond (expert questions)
- 32.1 %
- AIME (maths)
- 30.2 %
- LMArena · Coding
- 1362 · 87/221
- LMArena · Maths
- 1341 · 88/215
- LMArena · Creative writing
- 1305 · 84/224
- LMArena · Following instructions
- 1309 · 89/226
- LMArena · Hard prompts
- 1335 · 89/226
- LMArena · Long prompts
- 1326 · 83/205
- LMArena · Multi-turn dialogue
- 1343 · 78/224
- LMArena · Understanding images (Vision)
- 1142 · 26/49
Scores: LMArena, Epoch AI, EuroEval, Hugging Face, models.dev; as of 19 October 2026. Sources and method →
Use cases
Where the model makes sense.
Correspondence and texts in German
- Situation
- A company writes many letters, quotes and product texts and wants support without sending customer data to a cloud.
- What it looks like
- The model drafts texts following the company’s guidelines and examples; staff edit and send them.
- What to watch
- Provide your own wording and terms as examples, otherwise the texts sound generic.
Checking applications and forms
- Situation
- An authority or association receives many applications that must be checked for completeness.
- What it looks like
- The model reads the application, extracts the details and lists what is missing or unclear. A person makes the decision.
- What to watch
- Decisions about people are subject to strict rules under the EU AI Act and data protection law. The model prepares; it does not decide.
Website chat in your own data centre
- Situation
- Website visitors keep asking the same questions about services, opening hours or procedures.
- What it looks like
- The model answers based on your own pages and FAQ (RAG). It runs on your own hardware, and conversations go to no third party.
- What to watch
- Clearly state that an AI assistant is answering and refer to a person when in doubt.
Tenders requiring EU providers
- Situation
- Public tenders increasingly require a European provider or operation within the EU.
- What it looks like
- Mistral Small 3.2 runs in your own data centre or with a European host and comes from a European maker.
- What to watch
- What exactly is required (maker, operation, data location) is set out in the tender documents and must be checked per tender.
The use cases were written with AI assistance (Claude by Anthropic) and reviewed by vona. Scores come unchanged from the sources listed.
When another model fits better
In Epoch AI’s maths and expert-question tests it trails Qwen 3.8 27B. For calculations and logic, that model or gpt-oss-20b is a better fit.
How to start
ollama run mistral-small3.2With Ollama on a computer with 32 GB; for document search an interface such as AnythingLLM works well.
Suitable interfaces and programs from our directory: Ollama, LibreChat, AnythingLLM
Licence: Free for commercial use
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
Hugging Face ↗Similar models
Comparable models.
Mistral AI · France
Ministral 3 14B
Small Mistral AI model with image input for laptops and local assistants.
Free for commercial use · Runs on: Laptop with 16 GB
Use cases and details →Google · United States
Gemma 4 31B
Compact Google model with image input under the Apache 2.0 licence. It ranks near the top of EuroEval's German comparison and runs on a computer with 32 GB.
Free for commercial use · Runs on: Computer with 32 GB
Use cases and details →Mistral AI · France
Mistral Medium 3.5
Mid-sized Mistral AI model with image input and 262,000 tokens of context; quantised, it fits a Mac Studio with 128 GB. The licence excludes large companies.
Commercial use with conditions · Runs on: Mac Studio with 128 GB
Use cases and details →
FAQ
What hardware do I need for Mistral Small 3.2?
In the common 4-bit version, Mistral Small 3.2 needs about 16 GB of memory including headroom for context and runtime. That fits: Computer with 32 GB. The more accurate 8-bit version needs about 25 GB. Long documents increase the requirement.
May I use Mistral Small 3.2 commercially?
The licence allows commercial use, modification and redistribution; the main obligation is to keep the licence notice.
How good is Mistral Small 3.2 in German?
On LMArena, people rate Mistral Small 3.2 at 1340 points for prompts in German, rank 73 of 154 open models. The best way to see whether it suits your texts is a test with your own examples.
AI consulting
Your own language model for your company?
Together we find the model that fits your tasks, your hardware and your data protection requirements, test it with real documents and set up operation.