發生什麼事
Mistral handles the models and the physics modeling, led by Mistral Large 4, while NVIDIA's CUDA-X libraries and Warp provide the computing speed. The goal is a closed loop in which design changes, model training and simulation results feed directly into one another. SimReady assets and even 3D scenes are to be produced by models as well, with Omniverse libraries supplying the GPU acceleration (Mistral blog).
In Mistral's telling, the core of the deal is a set of public benchmarks plus open datasets. One use for the benchmarks is to test automated pipelines that convert CAD files into simulations; the other is to let everyone measure physics models with the same ruler, for both accuracy and speed (Mistral blog).
01Sources
- @MistralAI: Mistral's official account on X, posted 2026-10-21 19:35 (Taipei time), the earliest official announcement; the three areas of collaboration
- Mistral Advances AI for Engineering with NVIDIA: Mistral's official blog, published 2026-10-21 20:00 (Taipei time), after the X post, so this brief uses the X post time as the event time; Large 4 with CUDA-X and Warp, closed-loop process, Omniverse and SimReady, public benchmarks and CAD-to-simulation, CAE / EDA agent evaluations and target industries