tars vs llm — structured reasoning on sudoku
Tars is polished-snow's energy-based reasoning model. Instead of searching token by token, it settles into low-energy states across the full problem at once. This page compares tars against a large language model on hard Sudoku puzzles — a clean test of structured constraint satisfaction.