Beyond the two superpowers, the U.S. and China, European AI has finally delivered a respectable result. French AI company Mistral has released a public preview of its flagship model, Mistral Large 4. The community has nicknamed it “Le Chonk,” meaning “the big one.”
This “big one” really is big: it has about 1 trillion parameters in total and uses a mixture-of-experts architecture, with roughly 49 billion parameters activated for each inference. It was trained from scratch using 3,800 NVIDIA GPUs, with all training carried out at Mistral’s data centers in Europe. According to third-party benchmarks, it is currently the highest-scoring model outside the U.S. and China, with particularly strong performance on cybersecurity, finance, and legal tasks. It supports text and image inputs and has a context window of 512,000 to 1 million tokens.
The current model can already be used via the API at a discounted price, and the open-weight plans are scheduled to be released by the end of October. This is crucial: open weights mean enterprises can deploy the model on their own servers, with data kept local. This is highly attractive to European governments and companies that care about data sovereignty and compliance.
Why is this worth paying attention to? Over the past two years, frontier large language models have basically been monopolized by U.S. companies, while the strongest open-source players have largely come from China. Europe has long been worried about falling behind in the AI era. This time, Mistral has built a model at the trillion-parameter scale using relatively limited compute, proving that Europe still has competitiveness in AI, and providing a concrete example for the concept of “sovereign AI.”
My take: for investors, there are two implications. First, countries are building their own sovereign AI systems, which will continue to drive global demand for compute; chipmakers such as Nvidia are definite beneficiaries. Second, open-source models are getting stronger and will continue to push down the prices of model usage calls—this is pressure for companies that earn money by selling model APIs, but good news for application companies looking to reduce costs. The more intense competition at the model layer, the more likely value will flow toward both the compute and application ends.
The information above is for reference only and does not constitute investment advice.