Manifold solves the travelling salesman problem. Without being designed to

In a test originally posed in an academic paper, frontier LLMs (ChatGPT, Gemini, Claude) were challenged to solve routing problems on a grid map with random blockades and uncertain terrain. Every frontier model failed on the first 40-point grid.

Manifold engaged the same test, treating routing as a physics problem. The platform matched the performance of purpose-built optimization algorithms, scaling to a 1 million-point grid before test hardware ran out of memory. Manifold was never designed for route optimization.