RC RANDOM CHAOS

Mistral Unveils Le Chonk: A 1-Trillion-Parameter AI Powerhouse

· via Hacker News

Original source

Mistral Large 4: "Le Chonk"

Hacker News →

Mistral has launched a public preview of Mistral Large 4 (ML4), nicknamed ‘Le Chonk’, a 1-trillion-parameter multimodal model with 49 billion active parameters. ML4 excels in coding, agentic workflows, and multimodal understanding, outperforming other open-weight models and even some closed models in specific domains like visual grounding. The model is being red-teamed with cybersecurity leaders before its full release later this month. Trained on European infrastructure, ML4 emphasizes AI sovereignty, particularly in cybersecurity, where it can handle tasks blocked by closed models, such as vulnerability research and malware analysis. It is designed for self-deployment, giving organizations control over their AI capabilities.

ML4 demonstrates strong performance across various benchmarks, including cybersecurity, coding, and multimodal tasks. It scores highly on the Artificial Analysis Cyber Index and Cybench, showcasing its ability to identify and patch vulnerabilities. In coding, it outperforms competitors on several benchmarks and excels in agentic workflows, handling complex business tasks. Its multimodal capabilities allow it to analyze images, documents, and satellite imagery, making it useful in fields like engineering and disaster response. The model also shows strong scientific and mathematical reasoning, aiding in research and complex problem-solving.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.