Mistral AI launches Large 4 Le Chonk, a multimodal model for security & agentic workflows
Mistral AI has released Mistral Large 4, its new flagship open-weight multimodal model, following a public preview earlier this month. The model supports both text and images and is available for developers to download and customize, with additional reinforcement learning and fine-tuning bringing improvements to its coding and cybersecurity capabilities.
Large 4 focuses on defensive cybersecurity tasks such as inspecting system logs, auditing code, and identifying newly disclosed vulnerabilities, alongside offensive security workflows for finding weaknesses. It also targets agentic tasks involving planning, research, coding, and complex enterprise workflows, with Mistral claiming state-of-the-art performance on SWE-bench and Terminal-Bench.
The model also offers geospatial and visual grounding capabilities for analyzing maps, aerial and satellite imagery, and detecting events such as forest fires. Trained on more than 160 languages, including all official EU languages, Large 4 is available through Hugging Face, Mistral Studio, the Mistral API, OpenRouter, Vibe Code, and Nvidia-hosted endpoints.
Comments
The focus on defensive security work is what caught my eye here. Checking logs and auditing code are the kinds of unglamorous tasks where a capable open model could actually save teams a lot of time, and being able to adapt it for their own systems makes it more useful than a closed tool.