Mistral and NVIDIA
Mistral and NVIDIA team up to make engineering simulation faster
Mistral Large 4 · physics models · CUDA-X and Warp
Mistral is teaming up with NVIDIA on AI for engineering. Mistral Large 4 and Mistral’s physics models will run on NVIDIA’s CUDA-X libraries and Warp. The aim is faster simulation, with results fed straight back into design.
Physics AI in seconds
Physics AI predicts in seconds what a solver computes in hours
Traditional solvers stay for verification and edge cases
Mistral’s physics models learn from solver output, then predict how air flows or heat moves through a design in seconds, on one GPU. A traditional solver takes hours to weeks per design, and it still does the final verification.
Open physics benchmarks
Public benchmarks and open datasets for physics models
Mistral and NVIDIA will publish public benchmarks and open datasets for physics models. Researchers and engineering teams get one shared test for accuracy and performance, and a way to check workflows that turn a CAD model into a simulation.
Agent benchmarks for CAE and EDA
Benchmarks that score AI agents on real engineering tools
CAE and EDA · aerospace, automotive, semiconductors
They also plan benchmarks for AI agents in simulation and chip design software. An agent is scored on using the tools, finishing multistep work and getting a correct result, and on how long and how much compute it took.
Omniverse
Mistral models inside NVIDIA Omniverse
Mistral’s models will also be integrated with NVIDIA Omniverse libraries. That puts them to work generating 3D scenes on the GPU and creating SimReady assets for simulation.
Mistral Large 4 weights
Mistral Large 4 weights are planned for the end of October
The model at the center of this is Mistral Large 4, and its open weights are planned for the end of October. Until then, you can call the preview API in Mistral Studio. It has about a trillion parameters.
Nemotron Coalition, and a question
Next, an open frontier base model through the Nemotron Coalition
Which simulation or chip design tool would you want an open model tested in?
Mistral and NVIDIA are also building a frontier base model together, through the NVIDIA Nemotron Coalition, and those models will be open. Which simulation or chip design tool would you want an open model tested in?
