Mistral Medium 3.5
Mistral Medium 3.5 is Mistral AI’s mid-sized model with image input and 262,000 tokens of context. Quantised, it fits a Mac Studio with 128 GB. The licence allows commercial use but excludes companies with more than US$20 million in monthly revenue; for mid-sized businesses that is usually no obstacle.
Profile
- Maker
- Mistral AI · France
- Licence
- Commercial use with conditions
- Runs on
- Mac Studio with 128 GB
- Memory at 4 bits
- 86 GB
- Memory at 8 bits
- 133 GB
- Parameters
- 127.7bn
- Context
- 262,000 tokens
- Inputs
- text, image
- Released
- April 2026
Scores
- Human rating
- 1421 · rank 36 of 226 open
- In German
- 1453 · rank 11 of 154 open
- Epoch index
- 141.5 · rank 32 of 127 open
- LMArena · Coding
- 1461 · 24/221
- LMArena · Maths
- 1431 · 32/215
- LMArena · Creative writing
- 1374 · 45/224
- LMArena · Following instructions
- 1415 · 28/226
- LMArena · Hard prompts
- 1437 · 29/226
- LMArena · Long prompts
- 1416 · 38/205
- LMArena · Multi-turn dialogue
- 1424 · 33/224
- LMArena · Building websites (WebDev)
- 1264 · 53/58
- LMArena · Understanding images (Vision)
- 1223 · 18/49
Scores: LMArena, Epoch AI, EuroEval, Hugging Face, models.dev; as of 19 October 2026. Sources and method →
Use cases
Where the model makes sense.
Central assistant with a European model
- Situation
- A mid-sized company wants an AI assistant for everyone, with a model from Europe running in-house.
- What it looks like
- A Mac Studio runs the model, and LibreChat provides it to all staff with login via the company account.
- What to watch
- Keep an eye on the licence’s revenue threshold as you grow; above it, a licence from Mistral AI is required.
Texts in several European languages
- Situation
- An exporting company writes quotes, newsletters and product texts in German, French and other languages.
- What it looks like
- The model creates versions for other markets from a German original, and country managers review them.
- What to watch
- Have legal statements and product claims checked by experts in every language.
Analysing presentations and scans
- Situation
- Knowledge sits in old presentations and scanned documents that nobody searches any more.
- What it looks like
- The model reads slides and scans including graphics and creates summaries or answers questions about them.
- What to watch
- For poor scans, run text recognition first; it improves results.
Customer correspondence with many rules
- Situation
- Replies to customers must follow tone, legal notices and internal rules.
- What it looks like
- The model receives the rules as a fixed instruction and writes drafts that follow them.
- What to watch
- Regularly spot-check whether the rules are being followed.
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
Large companies above the revenue threshold should use Mistral Large 3 (Apache 2.0) or another model without this restriction. For a computer with 32 GB, Mistral Small 3.2 fits.
How to start
The weights are on Hugging Face; with a GGUF version it runs in LM Studio or llama.cpp on a Mac Studio with 128 GB.
Suitable interfaces and programs from our directory: Ollama, LibreChat, AnythingLLM
Licence: Commercial use with conditions
Commercial use is allowed but subject to conditions: Modified MIT licence: not permitted for companies with over US$20m monthly revenue; they need a commercial licence from Mistral AI.
Hugging Face ↗Similar models
Comparable models.
Mistral AI · France
Mistral Small 3.2
Compact Mistral AI model under the Apache 2.0 licence that runs on a computer with 32 GB. It ranks near the top of EuroEval's German comparison.
Free for commercial use · Runs on: Computer with 32 GB
Use cases and details →Alibaba (Qwen) · China
Qwen3.5 122B-A10B
Mid-sized Qwen model with image input and 262,000 tokens of context. With 4-bit quantisation it fits on a Mac Studio with 128 GB.
Free for commercial use · Runs on: Mac Studio with 128 GB
Use cases and details →OpenAI · United States
gpt-oss-120b
OpenAI's open text model from 2025 with adjustable reasoning effort and tool calls. Fits on a Mac Studio with 128 GB or a single 80 GB GPU.
Free for commercial use · Runs on: Mac Studio with 128 GB
Use cases and details →
FAQ
What hardware do I need for Mistral Medium 3.5?
In the common 4-bit version, Mistral Medium 3.5 needs about 86 GB of memory including headroom for context and runtime. That fits: Mac Studio with 128 GB. The more accurate 8-bit version needs about 133 GB. Long documents increase the requirement.
May I use Mistral Medium 3.5 commercially?
Commercial use is allowed but subject to conditions: Modified MIT licence: not permitted for companies with over US$20m monthly revenue; they need a commercial licence from Mistral AI.
How good is Mistral Medium 3.5 in German?
On LMArena, people rate Mistral Medium 3.5 at 1453 points for prompts in German, rank 11 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.