Mistral Large 4: 1T model trained on 4,000 Nvidia GPUs

Mistral is framing Mistral Large 4 around a third path. French president Macron described that positioning as "a third way in AI," set against closed models that can be unplugged and open models often made in China. The model, nicknamed Le Chonk for its 1 trillion parameters, is not open-weight yet; access runs through a public guardrail endpoint until the weights ship.

The training story is the differentiator Mistral is pushing. Mistral VP Science Pierre Stock said ML4 was trained entirely on Mistral's compute using only 4,000 Nvidia GPUs, two to three times less than its Chinese competitors and significantly less than the closed source competitors. Stock also noted that an open-weight model is easier to audit, a point aimed at enterprises and institutions where security concerns have been mounting. Mistral says it will work with trusted partners and governments so the open source weights can be used to defend, but not to perform malicious attacks.

With benchmarks still pending, Mistral's claim to best-in-class open-weight status rests on tests it has not yet published. The chip-design angle ties ML4 to two of its main backers — Dutch giant ASML, which led its Series C, and Samsung, which led its Series D last month at a €21 billion valuation. How much of that bet pays off is likely to hinge on whether the published benchmarks and the three-week weights release hold to plan.