Key facts
- Mistral AI released Mistral Large 4, a new large multimodal model with one trillion parameters.
- The model, nicknamed Le Chonk, is not yet an open-weight model but weights will be available in three weeks.
- Mistral plans to work with trusted partners and governments to ensure open-source weights are not used for malicious attacks.
- Mistral Large 4 was trained using 4,000 NVIDIA GPUs, which is significantly less than competitors.
- Mistral hopes ML4 will outperform closed models in specific areas like cybersecurity and finance.
- Mistral's backers include ASML and Samsung, which led its recent Series D funding round.
French AI lab Mistral AI has released Mistral Large 4, a new large multimodal model with one trillion parameters, aiming to compete with both closed and open-weight AI rivals. The model, nicknamed Le Chonk, is currently accessible via a public endpoint, with plans to release its weights in three weeks following safety testing.
Mistral VP Science Pierre Stock stated that the company will collaborate with trusted partners and governments to ensure the open-source weights are used defensively. He highlighted that an open-weight model is easier to audit, addressing security concerns from enterprises and institutions.
Mistral Large 4 was trained using 4,000 NVIDIA GPUs, a significantly smaller compute footprint compared to competitors. Mistral anticipates that ML4 will excel in specific areas such as cybersecurity, finance, and chip design, leveraging its multimodal capabilities to add value for customers.
The company's backers include Dutch giant ASML and Samsung, which recently led Mistral's Series D funding round at a valuation of €21 billion (approximately $24.39 billion). Mistral aims to be recognized as a frontier lab, not just an inference provider.