What is Mistral 3?

Mistral 3 is the next generation of open multimodal and multilingual AI models from Mistral. This release includes three state-of-the-art small, dense models (14B, 8B, and 3B parameters) and Mistral Large 3 – the most capable model to date – a sparse mixture-of-experts (MoE) model trained with 41B active and 675B total parameters. All models are released under the Apache 2.0 license, empowering developers with open access to frontier AI capabilities.

Who it's for

• Enterprise developers – building custom AI solutions that need strong multilingual support and image understanding without vendor lock-in
• Edge computing teams – deploying AI on local devices, robots, or resource-constrained environments where the Ministral 3 series offers the best performance-to-cost ratio
• Open-source AI community – researchers and tinkerers who want to fine-tune, customize, or self-host state-of-the-art models under a permissive license

Key features

Mistral Large 3: Frontier MoE architecture

Mistral Large 3 is Mistral's first mixture-of-experts model since the Mixtral series, trained from scratch on 3,000 NVIDIA H200 GPUs. It achieves parity with the best instruction-tuned open-weight models on general prompts while demonstrating image understanding and best-in-class performance on multilingual conversations.

Ministral 3: Edge-optimized intelligence

The Ministral 3 series comes in three sizes (3B, 8B, and 14B) with base, instruct, and reasoning variants – each with native multimodal and multilingual capabilities. These models deliver the best cost-to-performance ratio of any open-source model in their category.

Apache 2.0 licensing across all models

Every Mistral 3 model – from the smallest 3B variant to the massive Large 3 – is released under the Apache 2.0 license. This includes both base and instruction fine-tuned versions, providing a strong foundation for further customization across enterprise and developer communities.

Optimized for NVIDIA hardware and open-source tooling

Mistral partnered with NVIDIA, vLLM, and Red Hat to deliver efficient inference support. Large 3 runs on a single 8×A100 or 8×H100 node using vLLM, while Ministral models deploy seamlessly on DGX Spark, RTX PCs, laptops, and Jetson devices – from data center to robot.

What stands out

"Mistral Large 3 joins the ranks of frontier instruction-fine-tuned open-source models while the Ministral series offers the best performance-to-cost ratio in its category."

This dual release is remarkable because it covers the entire spectrum of AI deployment – from massive cloud-based MoE models to tiny edge-optimized dense models – all under one permissive license. Mistral Large 3 debuts at #2 in the OSS non-reasoning models category on the LMArena leaderboard, while the Ministral 3 series sets a new standard for cost-efficient local AI.

Worth checking out if…

You need open-weight models that balance performance, cost, and deployment flexibility. Mistral 3 is ideal if you're building multilingual applications, deploying AI on edge devices, or want a frontier MoE model you can customize freely. The Apache 2.0 license and broad hardware support make it a strong choice for both enterprise and community projects.