Mistral AI Releases Mistral Large 4 with 1 Trillion Parameters
Mistral AI has officially released the weights for Mistral Large 4, a 1-trillion parameter native multimodal model trained on NVIDIA Grace Blackwell GPUs. The model achieves a 48% score on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task compared to competing models. This release marks a significant milestone for European open-weight AI development, providing developers with a high-performance frontier model featuring high token efficiency for coding agents. It reinforces Mistral AI's position as a leading competitive force in the global AI landscape. Mistral Large 4 utilizes a Mixture-of-Experts architecture featuring 52 billion active parameters out of 1.05 trillion total parameters. It is licensed under a modified MIT license that allows free personal, academic, and general commercial use, while cloud service providers offering Model-as-a-Service must obtain a separate commercial license.
## BACKGROUND
Mistral AI is a prominent French artificial intelligence company founded in 2023 that focuses on developing open-weight large language models to promote European digital sovereignty. Mixture-of-Experts (MoE) is a machine learning architecture where only a subset of neural network parameters (active parameters) are activated for each token, allowing models to scale up total capacity while maintaining efficient inference costs.