Mistral and NVIDIA build engineering AI benchmarks
Mistral and NVIDIA plan public benchmarks and open datasets for physics and agentic AI in engineering, which could give aerospace, automotive and chip teams a common way to test models against real simulation and design work.
Mistral announced a collaboration with NVIDIA on what it calls "AI-native engineering workflows." The plan pairs Mistral Large 4 and Mistral's data-generation and physics-modeling pipelines with NVIDIA's CUDA-X libraries and NVIDIA Warp. The aim is faster simulation and closed-loop workflows that link simulation, model development and design. The companies will also connect Mistral's models to NVIDIA Omniverse libraries for 3D scene generation and SimReady asset creation.
The main deliverable is a set of evaluations. The companies plan public benchmarks and open datasets for measuring physics-model accuracy and performance, and for validating automated CAD-to-simulation workflows. A second set of benchmarks will cover agentic AI in computer-aided engineering (CAE) and electronic design automation (EDA). It will test whether agents can use engineering tools, finish multistep workflows and produce technically correct results, and it will also track execution time and compute cost.
"For too long, the pace of engineering has been set by the pace of simulation," said Neil Ashton, Mistral's VP of computational engineering. NVIDIA's Tim Costa said the shared benchmarks are meant "to build confidence in what AI can do."
The announcement gives no release dates, benchmark names, dataset sizes or results.
It also restates earlier news. Mistral Large 4, nicknamed "Le Chonk," is in public preview. It has roughly 1 trillion total parameters, 49 billion of them active, and was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral-run European data centers. Weights are planned for the end of October. The two companies are also working on a frontier base model through the NVIDIA Nemotron Coalition. Mistral says it is targeting 1 GW of compute capacity by 2030.