C-MCTD enables diffusion planners to generate plans 10× longer than training examples by systematically stitching together shorter plans through tree search.

Factorio Learning Environment is an evaluation for frontier models that offers exponentially scaling challenges.

Fast Monte Carlo Tree Diffusion (Fast-MCTD) achieves up to 100× speedup over MCTD through parallel rollouts and sparse trajectory planning, maintaining strong performance in complex long-horizon tasks while being computationally efficient.