tech
Introducing Mistral 3
Today, we announce Mistral 3, the next generation of Mistral models. Mistral 3 includes three state-of-the-art small, dense models (14B, 8B, and 3B) and Mistral Large 3 – our most capable model to date – a sparse mixture-of-experts trained with 41B active and 675B total parameters. All models are released under the Apache 2.0 license. Open-sourcing our models in a variety of compressed formats empowers the developer community and puts AI in people’s hands through distributed intelligence.
TL;DR
- Mistral AI launched Mistral 3, comprising three small, dense models (3B, 8B, 14B) and the flagship Mistral Large 3 (41B active, 675B total parameters).
- All models are released under the Apache 2.0 license, promoting open access and distributed intelligence.
- Mistral Large 3 is a sparse mixture-of-experts model trained on NVIDIA H200 GPUs, achieving parity with top open-weight models and demonstrating strong multimodal and multilingual capabilities.
- The Ministral 3 series offers the best cost-to-performance ratio, with variants optimized for edge and local use cases.
- Mistral AI collaborated with NVIDIA, vLLM, and Red Hat for hardware optimization, efficient inference, and accessibility.
- Models are available on various platforms including Mistral AI Studio, Hugging Face, Amazon Bedrock, and Azure Foundry.
- Custom model training services are offered for tailored enterprise AI solutions.
- Mistral 3 features frontier performance, multimodal and multilingual understanding, scalable efficiency, and agentic adaptability.