Mistral AI: the European developer of Apache-licensed open models

Contents

1. What it does for a company
2. Strengths and limits
3. Pricing
4. Questions to ask before choosing Mistral
5. Sources

Type	Open-weight model family
Headquarters	France
Pricing	Published
Website	docs.mistral.ai

Mistral AI is the European developer in this guide. Most of its open models, including Mistral Small 4 and the Ministral 3 family, are released under Apache 2.0, while its flagship, Mistral Large 4, comes under a licence of its own⁠12Models overviewMistral AI DocsMistral Ecosystem Open-Weight LicenseMistral AI (Hugging Face).

Best for: European companies that want a European developer and Apache-licensed models they can fine-tune.

Not the right fit for: Companies that want to offer Mistral’s flagship to customers as a service without a commercial licence.

What it does for a company

Mistral’s documentation lists Mistral Small 4, Ministral 3 in 3, 8 and 14 billion parameters, Voxtral Small and Shieldstral under Apache 2.0, and no model under its older research licence⁠1Models overviewMistral AI Docs. Mistral Small 4 has 119 billion parameters in total and activates 6.5 billion per token, with a 256,000-token context, which keeps the cost of serving it low for its size⁠3mistralai/Mistral-Small-4-119B-2603Mistral AI (Hugging Face).

The flagship, Mistral Large 4, was announced on 6 October 2026. It has about one trillion parameters, 52 billion active, a one-million-token context and image input⁠45Introducing Mistral Large 4Mistral AI · 6 Oct 2026mistralai/Mistral-Large-4-1T-A52BMistral AI (Hugging Face). Its licence, the Mistral Ecosystem Open-Weight License, is a modified MIT licence. Internal use is allowed; offering Large 4 to others as a model-as-a-service requires a separate commercial licence⁠2Mistral Ecosystem Open-Weight LicenseMistral AI (Hugging Face). Vendor-reported, the Large 4 preview scored 61.7% on DeepSWE v1.1⁠4Introducing Mistral Large 4Mistral AI · 6 Oct 2026.

Fine-tuning is better documented than for most families. The model cards link to examples for Axolotl, Unsloth and Hugging Face’s TRL; the TRL LoRA example for Large 4 runs on eight B300 GPUs, and Small 4 has its own Axolotl example⁠536mistralai/Mistral-Large-4-1T-A52BMistral AI (Hugging Face)mistralai/Mistral-Small-4-119B-2603Mistral AI (Hugging Face)TRL: Transformer Reinforcement LearningHugging Face.

Strengths and limits

Strengths

• A European developer with Apache-licensed models across sizes.
• Fine-tuning is documented: the Large 4 card links Axolotl, Unsloth and TRL examples, and the Small 4 card an Axolotl example.
• Small 4 activates only 6.5 billion parameters per token, which keeps serving costs low.

Limits

• Large 4 requires a separate commercial licence to operate it as a model-as-a-service; internal use is exempt.
• Weights only: training data is not released.

Choose another option if: You need the most permissive licence at flagship scale (DeepSeek) or a fully open model (OLMo).

Pricing

Weights are free for the Apache models. Commercial terms for Large 4 as a service are by agreement.

Questions to ask before choosing Mistral

• Is Small 4 enough for our work, or do we need Large 4, and if so, is our use internal?
• Which fine-tuning path fits our team: Axolotl, Unsloth or TRL?
• Will we self-host the weights, or use Mistral’s own platform?
• What is the commercial licence for Large 4 if we later offer it to customers?

See how Mistral compares with the other options in the main guide.

Sources

1. Models overview. Mistral AI Docs. Accessed 9 Nov 2026.
2. Mistral Ecosystem Open-Weight License. Mistral AI (Hugging Face). Accessed 9 Nov 2026.
3. mistralai/Mistral-Small-4-119B-2603. Mistral AI (Hugging Face). Accessed 9 Nov 2026.
4. Introducing Mistral Large 4. Mistral AI, 6 Oct 2026. Accessed 9 Nov 2026.
5. mistralai/Mistral-Large-4-1T-A52B. Mistral AI (Hugging Face). Accessed 9 Nov 2026.
6. TRL: Transformer Reinforcement Learning. Hugging Face. Accessed 9 Nov 2026.

Further reading

• Buyer's guide · Updated quarterlyThe best open-source AI in 2027, for companies that want to own itAn analysis of the open-weight model families a company can run today and the layer that turns one into its own, compared on seven published criteria, with sourced licences, costs and the established path from rented AI to owned AI.
• EuropeOpen-source AI in Europe: the models, the rules and where to run themEurope now has its own open-weight models, its own rules for them and its own reasons to keep AI close to home. A country-by-country guide to the models, the law and the path to an AI a European company controls.
• Practical guideHow to train your own AI model on company dataMost of the work in training a company model happens before any training starts. A step-by-step account of the data work, the choice of base model, the training methods and the rules that apply in Europe.

Also in: Deutsch · Français · Español · Italiano · Nederlands · Polski