Mistral and Nvidia team up on AI for engineering simulation, with open benchmarks planned

What was announced

Mistral and Nvidia said they will work together to speed up engineering simulation and build closed-loop workflows linking simulation, AI model training and product design.¹² Target industries include aerospace, automotive and semiconductors.²

The work combines Mistral Large 4 and Mistral’s physics modelling pipelines with Nvidia’s CUDA-X libraries, Warp and Omniverse, which will be used for 3D scene generation and simulation-ready assets.¹²

Open benchmarks for engineering AI

The companies plan public benchmarks and open datasets to compare physics models and to check automated CAD-to-simulation workflows.¹² Mistral says there will also be tests for AI agents in computer-aided engineering and electronic design automation, measuring tool use, multi-step task completion, technical correctness, time and computing cost.¹

Who said what

“For too long, the pace of engineering has been set by the pace of simulation,” said Neil Ashton, Mistral’s VP of computational engineering.¹ Nvidia’s Tim Costa spoke of building confidence “with shared industry benchmarks” in what AI can do.¹

Context

Mistral says Large 4 has about 1 trillion total parameters and 49 billion active, was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral-run European data centres, and is in public preview with weights due at the end of October.¹² Mistral targets 1 GW of computing capacity by 2030.¹² The two companies previously built Mistral NeMo together and are both part of Nvidia’s Nemotron Coalition.¹²

What it means

Engineering teams often wait hours or days for simulations, such as airflow over a wing or heat in a chip, before they can try the next design. AI models trained on simulation results can give fast approximate answers, letting engineers test more ideas and run the full simulation only on the best ones.

The planned benchmarks may matter most. Engineering firms need evidence before trusting AI with safety-critical design work, and shared, public tests make claims easier to compare. India has large engineering services and chip-design teams working for global companies, so open tools and benchmarks here could be widely used.

What to watch

• Release of Mistral Large 4 weights, planned for the end of October.
• Publication of the first engineering benchmarks and datasets.

Related: Mistral Large 4 in public preview and Nvidia’s open Vera Rubin rack designs.