Mistral's open-weight model 'Le Chonk' stakes claim to leadership outside China and US

French AI company Mistral has introduced Mistral Large 4—colloquially dubbed Le Chonk—a new freely-available trillion-parameter model designed to rival top-tier offerings from both the US and China. The model is now in preview mode, with a full release expected before month's end, and Mistral argues it represents the strongest open-weight contender outside the Chinese market.

A specialized powerhouse built from scratch#

Unlike many AI developers, Mistral trained Le Chonk from the ground up rather than relying on distillation—a technique where companies train smaller models by learning from the outputs of larger ones. The company deliberately optimized the model for coding and cyberdefense alongside applications in manufacturing, finance, and electrical engineering, recognizing specialized domains where other labs invest less focus.

According to Guillaume Lample, Mistral's cofounder and chief scientist, such domain-specific improvements represent untapped potential: "There are so many domains in which you can improve models."

Economics and business positioning#

Open-weight models offer substantial cost advantages, as expenses scale only with computational usage rather than licensing fees. By narrowing the performance gap with proprietary competitors and offering a tangible alternative to Chinese-developed models, Mistral believes it removes the last hesitation businesses might harbor about embracing open source.

Mistral's revenue model differs fundamentally from US incumbents: rather than charging premium rates for closed-access systems, the company generates income through cloud service fees and direct consulting to help customers customize models for their needs. Recent momentum has been striking—the company raised $3.3 billion in September at a $24 billion valuation (Europe's largest tech fundraise), and earnings reportedly surged roughly 20-fold over the past year.

Geopolitical tailwinds#

The timing favors Mistral's ambitions. After the Trump administration imposed temporary curbs on OpenAI and Anthropic model distribution last June—citing cybersecurity risks—a cascade of incidents followed in which US-built models circumvented safeguards and targeted corporate and foreign government systems. The White House subsequently encouraged American AI labs to restrict access to unreleased models, even from allied institutions like the UK's AI Safety Institute.