Mistral AI has announced Mistral 3, a significant advancement in their family of AI models. This release includes three state-of-the-art small, dense models (Mistral 3 14B, 8B, and 3B) and Mistral Large 3, their most capable model to date. Mistral Large 3 is a sparse mixture-of-experts model with 41B active and 675B total parameters, all released under the permissive Apache 2.0 license. This open-sourcing strategy aims to empower the developer community and democratize AI access.
Mistral 3 models are designed for high performance and cost-efficiency. Mistral Large 3, in particular, is positioned as a frontier open-weight model, trained on NVIDIA's H200 GPUs. It demonstrates strong capabilities in general prompts, image understanding, and excels in multilingual conversations, outperforming many existing open-weight models in these areas. It has achieved a high ranking on the LMArena leaderboard for OSS non-reasoning models.
The Ministral 3 series (3B, 8B, and 14B parameters) is optimized for edge and local use cases, offering base, instruct, and reasoning variants with multimodal capabilities. These models provide the best cost-to-performance ratio in their class, making advanced AI accessible for a wide range of applications, from consumer devices to enterprise solutions. The reasoning variants are particularly noteworthy for their accuracy on complex tasks, such as achieving 85% on AIME '25 with the 14B variant.
Mistral AI has collaborated with NVIDIA, vLLM, and Red Hat to ensure efficient deployment and accessibility. Mistral 3 models are optimized for NVIDIA hardware, including support for TensorRT-LLM and SGLang for low-precision execution. This collaboration enables efficient serving of long-context, high-throughput workloads and optimized deployments across various NVIDIA platforms, from data centers to edge devices.
Mistral 3 is available on Mistral AI Studio, major cloud platforms like Amazon Bedrock and Azure Foundry, and through various AI platforms such as Hugging Face, Modal, and Together AI. Mistral AI also offers custom model training services for organizations needing tailored AI solutions, ensuring maximum impact, security, and scalability.
