Mistral AI launches Mistral Large 4, its one-trillion-parameter multimodal model

Mistral AI has unveiled Mistral Large 4 (ML4), a multimodal AI model with one trillion parameters, marking its latest bid to challenge U.S. and Chinese dominance in artificial intelligence. Trained on 4,000 NVIDIA GPUs, the model is accessible via a guarded endpoint, with plans to release its weights in three weeks.

Mistral Large 4 enters the multimodal race

French AI lab Mistral AI has released Mistral Large 4 (ML4), a multimodal model with one trillion parameters. According to TechCrunch, the model is positioned to compete with both closed and open AI systems from U.S. and Chinese rivals. ML4, nicknamed Le Chonk, is currently accessible only through a public guardrail endpoint, with plans to release its weights in three weeks after completing safety testing.

Efficiency as a competitive edge

ML4 was trained entirely on Mistral’s infrastructure using 4,000 NVIDIA GPUs, a fraction of the hardware typically deployed by competitors. TechCrunch reports this is two to three times fewer than what Chinese rivals use and significantly less than closed-source developers. This efficiency could lower the barrier for advanced AI development, particularly for startups and research institutions with limited compute budgets.

Investors and use cases

Mistral AI raised its Series D last month, led by Samsung at a €21 billion valuation (~$24.39 billion). ASML, the Dutch semiconductor equipment manufacturer, previously led Mistral’s Series C. According to TechCrunch, ML4 is optimized for technical and enterprise applications, including cybersecurity, finance, and chip design—sectors where auditability and precision are critical.

Open weights pending safety review

While ML4 is currently accessible only through a guarded endpoint, Mistral plans to release its weights after safety testing. TechCrunch notes that open-weight models are easier to audit, a feature that could appeal to regulated industries. However, the three-week delay may temper immediate adoption among developers accustomed to full access.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

Mistral AI’s launch of Mistral Large 4 signals a strategic push to challenge dominant U.S. and Chinese AI models—not just through scale, but efficiency and transparency. By training a one-trillion-parameter model on just 4,000 NVIDIA GPUs, Mistral is demonstrating that advanced AI development isn’t solely the domain of those with massive compute budgets. This could pressure competitors to rethink their own efficiency trade-offs, especially as hardware costs remain a major barrier to entry.

The decision to release ML4’s weights after safety testing is a balancing act. On one hand, open-weight models are easier to audit and build upon, which could accelerate adoption in regulated industries like finance and cybersecurity. On the other, delaying the release of weights—while understandable from a safety standpoint—risks dampening the immediate impact among developers who prioritize full access. For founders and operators, this launch underscores the growing importance of multimodal capabilities and cost-efficient training as differentiators in a crowded AI market.

Key takeaways

• Mistral AI has released Mistral Large 4 (ML4), a multimodal model with one trillion parameters, targeting global AI leadership.
• ML4 is currently accessible via a guarded endpoint, with open weights planned for release in three weeks.
• The model was trained on 4,000 NVIDIA GPUs, a fraction of what competitors typically use.
• Mistral’s latest valuation stands at €21 billion after a Samsung-led Series D.
• Optimized use cases include cybersecurity, finance, and chip design, reflecting Mistral’s focus on enterprise and technical applications.

FAQ

What is Mistral Large 4?

Mistral Large 4 (ML4) is a multimodal AI model developed by Mistral AI, featuring one trillion parameters. It is designed to compete with both closed and open AI models from U.S. and Chinese companies.

How was Mistral Large 4 trained?

ML4 was trained using 4,000 NVIDIA GPUs, significantly fewer than what competitors typically use. This highlights Mistral’s focus on efficiency and cost-effective AI development.

When will the weights for Mistral Large 4 be available?

Mistral plans to release the weights of ML4 in three weeks, pending the completion of safety testing.

What are the key use cases for Mistral Large 4?

ML4 is optimized for applications in cybersecurity, finance, and chip design, reflecting its potential for technical and enterprise use cases.

Who are Mistral AI’s key investors?

Mistral AI’s Series D was led by Samsung at a €21 billion valuation. ASML, a Dutch semiconductor equipment manufacturer, also backed the company in an earlier funding round.

Related on Lazyfounder

• Samsung
• More ai news
• PearX demo day highlights 5 startups in AI, semiconductors, and privacy