French AI lab Mistral AI on Tuesday announced Mistral Large 4 (ML4), a new large multimodal model intended to compete with both American and Chinese offerings. The company’s development has been described by French President Emmanuel Macron as a “third way” in AI.
Nicknamed Le Chonk, ML4 contains roughly 1 trillion parameters. It is not yet distributed as an open‑weight model: access is currently provided only through a public but guarded endpoint. Mistral plans to release the model weights in three weeks, after completing safety testing.
Pierre Stock, Mistral’s VP Science, told TechCrunch: “In the meantime, we’ll work with trusted partners and governments to make sure that the open source weights can be used to defend, but not to [perform] malicious attacks.”
Security concerns have increased in recent months, particularly among Mistral’s core customers — enterprises and institutions. Stock noted that an open‑weight model is easier to audit, which is an important consideration for those clients.
A less visible but significant point is that ML4 was trained entirely on Mistral’s own compute. According to the company, the training used 4,000 Nvidia GPUs — roughly two to three times fewer than their Chinese competitors and substantially fewer than some closed‑source rivals, per Stock.
Benchmark results are not yet available, but Mistral hopes ML4 will be best in class among open‑weight models, especially outside China. Stock added that focused training could allow ML4 to outperform closed models in specific tasks that matter to the company’s customers, where multimodal capabilities provide extra value.
Stock identified cybersecurity and finance as optimized use cases for ML4, and also highlighted chip design. The latter is strategically important to two of Mistral’s main backers: Dutch semiconductor equipment maker ASML, which led the company’s Series C round, and Samsung, which led the Series D round closed last month when Mistral reached a €21 billion valuation (about $24.39 billion).
At the time of the Series D, Mistral emphasized that hosting Chinese models did not mean the company was pivoting to become merely an inference provider. With Le Chonk now introduced, Mistral expects to continue being regarded as a frontier lab.
Why this matters
The announcement highlights European efforts to offer a competitive alternative to U.S. and Chinese AI ecosystems while attempting to balance openness with safety. The planned public release of weights after a three‑week safety period and collaboration with partners and governments illustrate Mistral’s cautious approach to mitigating potential misuse of an open‑weight, large multimodal model.
