Mistral models

Mistral AI's open-weight models from France: Apache 2.0 small models, the 128B Mistral Medium 3.5 and the 1T-parameter Mistral Large 4.

Open weights
Models 8
Context up to 256K
Updated Nov 2026

Hugging Face Compare

On this page

At a glance
Models & versions
VRAM & download size
Run it locally
Best uses
Strengths & limits
About Mistral
Guides
FAQ
Compare

At a glance

Type: Open-weight models
Developer: Mistral AI (France)
First release: Sep 2023
Models tracked: 8 (4 current)
Longest context: 256K tokens
Abilities: Tool callingReasoning modeImage input
Licence: Small, Ministral and older models are Apache 2.0. Mistral Medium 3.5 uses a modified MIT licence and Mistral Large 4 the Mistral Ecosystem Open-Weight License; check both before commercial use.
Smallest local option: Ministral 3 3B · ~4 GB
Website: mistral.ai
Hugging Face: huggingface.co/mistralai

Newest model

Mistral Large 4

Released Sep 2026

Smallest option

Ministral 3 3B

~4 GB VRAM · 3 GB download

Mistral models and versions

Model	Released	Sizes	Context	Abilities	Licence	Status
Mistral Large 4 Largest Mistral model, multimodal, 1M context	Sep 2026	1T Too large for most servers · Ollama cloud	—	ToolsReasoningVision	Mistral Ecosystem Open-Weight License	Current
Mistral Medium 3.5 One model for chat, reasoning and coding	Mar 2026	128B	256K	ToolsReasoningVision	Modified MIT	Current
Ministral 3 Edge models with vision and 256K context	Dec 2025	3B · 8B · 14B 3B, 8B, 14B	256K	ToolsVision	Apache 2.0	Current
Mistral Small 3.2 Reliable 24B model with vision and tool calling	Jun 2025	24B	128K	ToolsVision	Apache 2.0	Current
Magistral Small Reasoning model, replaced by Medium 3.5 in Le Chat	Jun 2025	24B	39K	ToolsReasoning	Apache 2.0	Previous generation
Mistral NeMo 12B model built with NVIDIA	Jul 2024	12B	128K	Tools	Apache 2.0	Legacy
Mistral 7B The classic small open model	May 2024	7B	32K	Tools	Apache 2.0	Legacy
Mixtral Early open MoE models	Dec 2023	8x7B · 8x22B 8x7B, 8x22B	64K	Tools	Apache 2.0	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 Mistral 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.

Model	4-bit (default)	8-bit	16-bit	VRAM (4-bit)	Fits on
Mistral Medium 3.5 128B	80 GB	138 GB	255 GB	~88.5 GB	128 GB GPU
Ministral 3 3B	3 GB	4.5 GB	7.7 GB	~4 GB	8 GB GPU
Ministral 3 8B	6 GB	9.9 GB	18 GB	~7.5 GB	8 GB GPU
Ministral 3 14B	9.1 GB	15 GB	28 GB	~11 GB	12 GB GPU
Mistral Small 3.2 24B	15 GB	26 GB	48 GB	~17 GB	24 GB GPU
Magistral Small 24B	14 GB	25 GB	47 GB	~16 GB	16 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 Mistral Medium 3.5 locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run mistral-medium-3.5:128b

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

# Ollama serves an OpenAI-compatible API on port 11434
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "mistral-medium-3.5:128b", "messages": [{"role": "user", "content": "Hello"}]}'

# Full-precision weights from Hugging Face, OpenAI-compatible API on :8000
vllm serve mistralai/Mistral-Medium-3.5-128B

vLLM loads the original 16-bit weights, so it needs roughly the 16-bit size above in GPU memory, plus room for the context. See Install vLLM on Ubuntu 26.04.

What Mistral is good for

European data residency

A French vendor with open weights you can host in your own EU data centre.

Instruction following on one GPU

Mistral Small 3.2 (24B) follows formats and function calls reliably on 16–24 GB.

Edge and mobile

Ministral 3 (3B–14B) brings vision and a 256K context to small hardware.

Strengths and limits

Strengths

Apache 2.0 on all small and mid-size models.
Good multilingual quality in European languages.
Clean, consistent tool calling and JSON output.

Watch out for

Mistral Large 4 and Medium 3.5 use custom licences.
Medium 3.5 (128B dense) needs about 90 GB at 4-bit.

About Mistral

Mistral AI was founded in Paris in 2023 by former Google DeepMind and Meta researchers. Mistral 7B (September 2023) and Mixtral 8x7B (December 2023) were among the first open models to beat larger rivals, and Mixtral popularised mixture-of-experts for open weights.

Today’s line-up has three tiers. Ministral 3 (3B, 8B, 14B) is for edge devices, and Mistral Small 3.2 (24B) is the workhorse for a single GPU; both are Apache 2.0 with image input. Mistral Medium 3.5 (March 2026) is a 128B dense model that merges instruction following, reasoning and coding; it replaced Magistral and Devstral 2 in Mistral’s own products. Mistral Large 4 (September 2026) is a 1-trillion-parameter multimodal MoE with a 1M-token context.

Mistral’s coding models are covered on the Devstral page.

Guides for running Mistral

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 Mistral model?

The newest Mistral model is Mistral Large 4, released in Sep 2026.

How much VRAM do I need to run Mistral locally?

The smallest option, Ministral 3 3B, is a 3 GB download at 4-bit and needs about 4 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 Mistral with Ollama?

Install Ollama, then run "ollama run ministral-3:3b". 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 Mistral commercially?

Small, Ministral and older models are Apache 2.0. Mistral Medium 3.5 uses a modified MIT licence and Mistral Large 4 the Mistral Ecosystem Open-Weight License; check both before commercial use. Always read the licence of the exact model you deploy; it is linked in the models table.

Compare Mistral

Mistral vs Llama

← All AI models