Mistral AI Unveils New Generation of Open-Source Models: From Ultra-large to Lightweight
Table of Contents
Published: april 29, 2024 | Last Updated: April 29, 2024
New Models Released Under Apache 2.0 License
Mistral AI has announced the release of its latest suite of large language models, including the flagship ‘Mistral Large 3’ and a series of smaller, dense models - 14B, 8B, and 3B parameters. Crucially, all models are available to the developer community under the permissive Apache 2.0 license, enabling free use and modification.
Mistral Large 3: A Powerful Mixture-of-Experts Model
The core of this release, Mistral Large 3, is a sparse mixture-of-experts (SMoE) model boasting 41 billion active parameters from a total of 675 billion parameters. According to Mistral AI, the model was trained from scratch utilizing a substantial infrastructure of 3,000 NVIDIA H200 gpus.
Performance benchmarks, as reported by Mistral AI, place the model 2nd in the open-source non-inferential model category and 6th overall among all open-source models on the LM Arena leaderboard. The company claims it achieves performance comparable to leading open-weight models in general tasks, and excels in image understanding and multilingual capabilities beyond English and Chinese.
Ministral 3 Series: Optimized for Edge and Local Deployment
Complementing the flagship model, the ministral 3 series (3B, 8B, 14B) is designed for deployment in resource-constrained environments, such as edge devices and local applications. Each model size is offered in Base, command-tuned, and inference versions. All models within the series incorporate image understanding and multilingual support.
The inference conversion models within the Ministral 3 series are specifically engineered for complex problem-solving through extended reasoning. Notably, Mistral AI reports that the 14B model successfully solved 85% of the problems presented in the AIME 2025 mathematics competition.
I hope you found this article insightful! if you enjoyed learning about Mistral AI’s new models, please share it with your network. I’d love to hear your thoughts in the comments below, and don’t forget to subscribe to World Today News for the latest updates in AI and technology.
