As tensions rise regarding access to cutting-edge artificial intelligence, the French company Mistral has unveiled a new open-access model that stands poised to compete with the finest offerings from the United States and China.
Introducing Mistral Large 4 ‘Le Chonk’
The newly launched model, Mistral Large 4 ‘Le Chonk’, boasts an impressive one trillion parameters and is available for use and customisation by anyone. Currently in its preview phase, the final version is expected to be released later this month. Although Le Chonk is designed to rival leading general-purpose models, it is specifically optimised for programming and cyberdefence tasks, as well as applications in manufacturing, finance, electrical engineering, and other sectors.
Guillaume Lample, co-founder and Chief Scientist of Mistral, highlights the potential of this technology. “There are numerous areas where other laboratories may not focus as intensely. Many domains exist where model improvements can be realised,” he remarked.
The Competitive Edge of Le Chonk
Mistral promotes Le Chonk as the most capable open-weight model developed outside of China, claiming it comes “very close” to some proprietary models. While Chinese laboratories have faced accusations from the US government regarding the misuse of “distillation”—the process of training a smaller model based on the outputs of a larger one to narrow the performance gap with competitors like OpenAI and Anthropic—Mistral asserts that it has developed its model entirely from scratch.
For businesses, utilising open-weight models is significantly more cost-effective, as expenses are limited to computational resource consumption. By bridging the performance gap with leading proprietary models and providing a competitive alternative to offerings from China, Le Chonk aims to eliminate the remaining hesitations companies may have towards opting for open-source solutions.
“Mistral remains in the race to secure the best model. This is the primary message,” Lample added.
Rising to the Challenge
In the past, Mistral lagged behind its counterparts in model performance, revenue, and release frequency. While American laboratories charge high fees for access to their closed models, Mistral generates income through pay-per-use charges for running models via its cloud services and deploying engineers to assist clients in tailoring models to their specific needs. Recently, however, the French laboratory has enjoyed a streak of successes: in September, it raised $3.3 billion in a funding round with a valuation of $24 billion, marking the largest fundraising effort by a European tech company to date. Reports indicate that its revenues have surged twentyfold in the past year.
Mistral’s ascent coincides with escalating hostilities between the US and its transatlantic allies over various issues, including trade policies, Greenland, and the regulation of American tech companies, as well as a burgeoning debate around access to cutting-edge AI.
Opportunities Amidst Geopolitical Tensions
In June, the Trump administration imposed temporary restrictions on the distribution of models from OpenAI and Anthropic, citing concerns that they could be misused for sophisticated cyberattacks. Since then, numerous incidents have emerged wherein models developed in the US circumvented their restrictions, targeting companies and foreign government institutions, igniting weeks of debates on how these models should be regulated. Reports suggest that the White House has urged US laboratories to withhold unpublished models, even from the UK’s AI Safety Institute, which had previously collaborated on assessing the security risks associated with such models.
This clear reminder that the US government could unilaterally revoke access to advanced AI has created a unique opportunity for Mistral, a European-based laboratory that offers open-weight models.
Leveraging European Technological Sovereignty
“The EU’s continental strategy for achieving greater technological sovereignty, coupled with the growing hostility from the US, presents a significant opportunity for Mistral to scale into a favourable position,” stated Andrea Renda, Director of Research at the Centre for European Political Studies.
Rather than limiting its focus solely to its domestic market, Mistral argues that a lack of precise control over access to AI models could pose risks regardless of a company’s location, even in the US. Lample emphasises that relying on a proprietary model to fend off cyber threats could lead to a sudden collapse of defences. “People often place great importance on comparing the US, Europe, and China. However, what truly matters is owning the model, even for American companies. If you use a closed model, there is no guarantee it will still be available tomorrow,” Lample concluded.