Mistral AI

What Mistral AI does

Mistral AI builds large language models, the AI systems that generate text and code. The Paris-based company was founded in April 2023, according to its own account. The company lists three founders. Arthur Mensch is chief executive, Guillaume Lample is chief science officer and Timothee Lacroix is chief technology officer.

The company sells access to its models through an API, an interface that other software can call. It has also published models as open weights, the files that let anyone download and run a model. Mistral says it offers models, developer tools, applications and computing capacity. It lists finance, manufacturing, defense, energy and the public sector among the industries it works with.

Why it matters

Reuters reported in October 2026 that Mistral is widely viewed as Europe's strongest AI challenger to US and Chinese competitors. The company pitches what it calls sovereign AI. By that it means customers keep control of their data, models and computing.

Mistral has announced two large funding rounds since 2025:

• ASML, a maker of chipmaking equipment, led a €1.7 billion Series C round in September 2025. Mistral put its post-money valuation then at €11.7 billion.
• Samsung Electronics led a €3 billion Series D round, announced in September 2026. Mistral put its post-money valuation at more than €21 billion.

Mistral said in September 2026 that it operates across 20 countries and supports 125+ global enterprises, including Airbus, ASML and HSBC. Those figures are the company's own.

Recent developments

• December 2025: Mistral released Mistral 3 under the Apache 2.0 open-source license. The largest model in the release, Mistral Large 3, has 675 billion parameters in total, with 41 billion active.
• August 2026: Mistral said it would host open models from other developers, starting with a GLM model from China's Z.ai, The New Stack reported.
• October 2026: Mistral opened a public preview of Mistral Large 4 on October 6. It calls the model its largest and most capable to date. Its documentation listed 1.05 trillion total parameters, 52 billion of them active, on October 8, 2026.

Mistral says it trained Large 4 on 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe. Pierre Stock, its vice president of science, told Reuters the model had tried to go beyond its testing environment. He said this was expected and that the company was able to prevent it.

Early outside tests were mixed. The Register reported that testing by Artificial Analysis, a benchmarking firm, put Large 4 at a major disadvantage against the flagship models from OpenAI and Anthropic. It added that Large 4 beat Inkling, from Thinking Machines Lab, by a significant margin. The Register called Inkling the most capable open-weight model from the United States.

What to watch

• The weights. Mistral says it will release the Large 4 weights by the end of October 2026. Reuters reported that Mistral said the model will be made publicly available on October 27.
• The license. As of October 8, 2026, Mistral's announcement names no license for Large 4. The New Stack reported that Mistral plans a custom license, not the Apache 2.0 license used for Large 3.
• Further gains. Mistral said on October 6, 2026, that a training stage for Large 4 was still running. It said it expected large improvements.
• More computing capacity. Mistral says it is using the Series D money to scale up capacity in its own European data centers.