Mistral Launches ML4, a 1 Trillion-Parameter AI Model with Advanced Capabilities

Mistral has launched a public preview of its new AI model, Mistral Large 4 (ML4), which has 1 trillion parameters and 49 billion active parameters.

ML4 is designed for open-weight performance and is capable of handling complex tasks in coding, cybersecurity, finance, and multimodal understanding.

The model was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's European data centers and will be available for public use by the end of October 2026.

ML4 has demonstrated state-of-the-art performance in critical enterprise workloads, outperforming both open-source and closed models in various benchmarks, particularly in cybersecurity tasks.

The model's training data includes multilingual sources, covering over 160 languages, and it is designed to provide organizations with autonomy in deploying AI for security operations.

Mistral has collaborated with leading enterprises across multiple industries to develop ML4, which will serve as the foundation for future specialized models.

The model's capabilities include advanced visual grounding, scientific problem-solving, and high performance in legal and financial tasks, making it a versatile tool for professionals.

ML4 has achieved high scores in various benchmarks, including the Artificial Analysis Cyber Index and Cybench, indicating its effectiveness in identifying and addressing security vulnerabilities.

The company plans to release further details on ML4's architecture and benchmarks as it continues to refine the model and expand its capabilities.