Mistral Large 3 is a state-of-the-art open model, trained from scratch on 3000 NVIDIA H200 GPUs, and represents a substantial step forward in pretraining. It achieves parity with the best instruction-tuned open-weight models on general prompts, while also demonstrating image understanding and best-in-class performance on multilingual conversations. The model debuts at #2 in the OSS non-reasoning models category, providing a strong foundation for further customization across the enterprise and developer communities.
Mistral 3 offers a family of models built for frontier intelligence, multimodal flexibility, and unmatched customization. The models are available in various sizes, from 3B to 675B parameters, and can be deployed for coding, creative collaboration, document analysis, or tool-use workflows with precision. The Ministral 3 series achieves the best cost-to-performance ratio of any OSS model, making it suitable for edge and local use cases, and can be customized to meet specific needs through fine-tuning or enterprise-grade training.
