Can Mistral Large 4 Close Europe’s AI Gap?
The Large 4 model from French AI company Mistral is designed to compete with the world’s most capable open models. For one of Europe’s leading AI companies, that is strategically important. Mistral has unveiled its latest flagship model, Large 4. According to the company, the model offers particularly strong capabilities in programming and cybersecurity and is the most powerful open model developed in Europe and the US to date. Mistral says it closes the gap with the leading open-source models, including those developed in China.
Pressure on Mistral has increased significantly in recent months. The French AI company’s previous models, Large 3 and 3.5, had fallen behind competitors from China and the US. Large 3.5 was based on what was increasingly considered an outdated Meta architecture and was more expensive than stronger competing models from Chinese provider DeepSeek. Industry advisers have also reported that demand for Mistral models has declined significantly in recent months compared with two years ago.
ML4, as the new model is abbreviated, is intended to change that and make Mistral competitive again in the race to develop leading AI technology.
For Europe, this is strategically important. Mistral is one of the few European AI providers able to compete internationally at the level of the most capable models. For companies and governments in particular, a powerful European model could eventually provide an alternative to the major American and Chinese providers.
During its latest funding round in September, Mistral said that the €3 billion raised would primarily be invested in model development and the expansion of computing capacity. The French AI lab aims to remain competitive over the long term and offer a powerful alternative for enterprise customers that want to operate models themselves.
Adapted To Customer Needs
According to Mistral, Large 4 offers state-of-the-art capabilities in areas that are particularly important to customers. These include cybersecurity, manufacturing and financial services. The model is also multimodal and can process images as well as text.
The company has also said that Large 4 offers significantly improved agentic capabilities. It can work on complex tasks for hours and interact with different tools without losing context. Earlier generations of the model were not able to do this. According to Mistral, ML4 could therefore take over work that currently requires specialists several hours or even days to complete.
Mistral has won major industrial customers including ASML, Samsung and Airbus in recent months and works with governments including Luxembourg and France. The AI company says it has adapted the model’s capabilities to the needs of its customers.
Alongside Luxembourgish and Korean, ML4 was trained on more than 160 languages, according to Mistral.
Whether ML4 can meet expectations will become clearer after extensive testing by independent experts in the coming weeks. Mistral initially released a preview version of the model through an API on Tuesday. The model weights, which will allow developers around the world to download, run and customise the model, are scheduled to follow on 27 October.
Following the release of the model weights, Mistral is expected to move from 15th to sixth place in the intelligence index compiled by benchmarking platform Artificial Analysis. That would place ML4 behind only models developed by Chinese providers.
Large 4 Is Designed To Handle Complex Programming Tasks
Programming and cybersecurity are intended to be particular strengths of the model. Mistral says its development team focused heavily on agentic programming capabilities. ML4 is designed to analyse existing software projects or independently add new functions. According to Mistral, the model closes the gap with the leading models in programming performance. The company also says the model performs particularly well on tasks related to semiconductor development and testing.
