French company Mistral announced its new Mistral Large 4 model, nicknamed Le Chonk, which it says is capable of competing with the best closed artificial intelligence models from the United States and China. The model contains one trillion parameters and is currently available in a preview version with open weights that allow companies and developers to use and customize it, with the final version scheduled to be released before the end of October 2026.
A General Model with a Sectoral Focus
Le Chonk was designed to compete with leading general-purpose models, but it particularly targets programming and cyber defense, in addition to applications in manufacturing, finance, electrical engineering, and other specialized fields. Mistral believes these sectors give it room to stand out because major artificial intelligence labs do not focus equally on all professional fields.
The company presents the model as the most powerful open-weight model developed by an entity outside China, and says it comes very close to the level of some closed commercial models. It also emphasizes that it was trained from scratch, distinguishing it from criticism directed at Chinese laboratories over the use of distillation techniques to train smaller models based on the outputs of larger models.
What Changes in Practice?
The most significant impact is not related only to the model's size, but also to making its weights available for operation and customization. The cost of running open models is primarily tied to the computing capacity consumed, rather than access fees for a closed model charged by an external provider. Narrowing the performance gap with commercial models could make choosing an open model more attractive to companies that need to control their data or adapt the model to their operations.
However, the source does not provide independent test results or detailed comparative metrics proving the claim of equivalence with the best closed models. The available version is also still experimental, so the quality of the final release and the operational requirements of running a model of this size remain open questions.
The Competition for Sovereignty and Access
The move comes at a time of growing concern about institutions relying on models whose availability could change at the decision of a provider or government. The article stated that U.S. President Donald Trump's administration imposed temporary restrictions in June on the distribution of models from OpenAI and Anthropic, due to concerns that they could be used to launch advanced cyberattacks. Reports also emerged about a request to block undisclosed models even from the United Kingdom's AI Safety Institute.
This context gives Mistral a strategic opportunity in Europe, particularly as the European Union seeks to strengthen technological sovereignty and transatlantic relations grow more tense. The company is also drawing on its commercial expansion; it raised $3.3 billion in September in a round that valued it at $24 billion, while its revenue was reportedly nearly 20 times higher over the past year.
certi.news Analysis
If the final version maintains the announced performance level, Le Chonk could become an important option for institutions seeking an advanced model with greater control and customization capabilities, particularly in programming, cybersecurity, and engineering sectors. Nevertheless, open weights do not eliminate the challenges of infrastructure, operation, and governance, nor do they alone prove that the model actually outperforms closed alternatives. Its practical value will be determined once independent tests emerge, along with clearer details about licensing, deployment requirements, and the model's performance on specialized tasks.