NVIDIA cuOpt™ is intended to solve routing and delivery problems, with a wide range of constraints and features, with breakthrough speed and world-class accuracy. The NVIDIA cuOpt GPU-accelerated solver delivers high-quality solutions quickly and reliably and can scale to large problems (thousands of delivery locations per invocation).
Various constraints and problem variants are supported, including over 700 variants of the basic Traveling Salesperson Problem (TSP). Supported features include pickup-and-delivery, heterogeneous fleets which may or may not start or end at the depot, multi-depot, and task tag matching to vehicle types.

The cuOpt API is directly accessible via Python or can be wrapped as a service with a RESTful API for a much simpler delivery option. The cuOpt production-ready solver is offered as a RESTful API Service that is simple and easy to use with any browser. Using the NVIDIA cuOpt service, users do not need to worry about resource provisioning or environment setup; simply pip install the cuOpt Python thin client, enter your credentials, and solve your routing problems.