French AI lab Mistral AI releases Mistral Large 4, a one-trillion-parameter multimodal model trained on 4,000 GPUs.

French artificial intelligence lab Mistral AI released a new large multimodal model called Mistral Large 4 on Tuesday. Nicknamed Le Chonk, the model features one trillion parameters. It aims to compete with American and Chinese rivals as an alternative to closed and open models. The model is currently accessible through a public guardrail endpoint.

Mistral plans to release its open-source weights in three weeks after safety testing finishes. Pierre Stock, Mistral’s vice president of science, stated that the company trained the model using only 4,000 NVIDIA graphics processing units. This hardware usage is significantly lower than closed-source competitors and two to three times less than Chinese competitors.

The company optimized the system for specific use cases, including cybersecurity, finance, and chip design. Chip design is a core area for two of Mistral's main backers, ASML and Samsung. Samsung led a funding round last month that valued the company at €21 billion, or about $24.39 billion. Benchmark results for the model are still pending.