The French company is trying to catch up with international competition from the USA and China with its new flagship model.
AI-generated summary
Mistral is struggling with falling demand and a technological lag compared to models from the USA and China.
Dusseldorf. Mistral has introduced its latest flagship model, Large 4. According to the company, the model is said to have particularly good programming and cybersecurity capabilities and is the most powerful open model from Europe and the USA to date. Chief scientist Guillaume Lample said in advance in a press session: “It closes the gap with the best open source models, including all the Chinese ones.”
The pressure on Mistral has increased significantly in recent months. The French AI company's previous models, Large 3 and 3.5, were left behind by competition from China and the USA. AI researcher Niels Rogge wrote on LinkedIn that Large 3.5 is based on an “outdated architecture” from Meta and is more expensive than the stronger competition from the Chinese provider Deepseek. Consultants also reported to the Handelsblatt that Mistral models were in demand significantly less in the past few months than two years ago.

The book industry is increasingly defending itself against AI providers who use copyrighted works to train their models without permission. While publishers like Penguin Random House are suing, the case of author Thélyson Orélien is causing a stir.

Sam Altman, head of OpenAI, spoke to the UN Security Council in New York about the risks of artificial intelligence. He warned of a loss of control of technology and a dangerous concentration of power in a few hands.

In just 45 seconds, an unknown person manipulates a Tesla wallbox using a laptop. The malware then infects parked cars and from there spreads to other charging stations.

Kleinanzeigen.de analyzes chat messages between sellers and interested parties using AI to generate automated response suggestions. The feature is enabled by default, but can be disabled in Settings. The company emphasizes that personal data will not be used for AI training.

US President Donald Trump has introduced the AI platform America.gov, which is intended to connect citizens with US authorities via an AI agent. The platform is intended to provide access to 29,000 government websites and handle tasks such as applications, but experts warn of hallucinations and privacy concerns, particularly for immigrants and welfare.

In an attack on the Danish Central Register of Persons, unknown persons gained access to data on 8.8 million people. The legitimate access of a private company was used.