This paper presents a novel approach to planning chemical syntheses using Monte Carlo Tree Search (MCTS) combined with deep neural networks. The method, called 3N-MCTS, uses three neural networks: an expansion policy network, a rollout policy network, and an in-scope filter network. The expansion policy network guides the search by suggesting possible transformations, the rollout policy network predicts the best transformations, and the in-scope filter network ensures that only feasible reactions are considered. The system was trained on 12 million reactions, representing essentially all reactions ever published in organic chemistry. The method outperforms traditional search methods in terms of speed and the number of molecules solved. It can solve almost twice as many molecules and is 30 times faster than traditional methods. The system can generate routes that are indistinguishable from those found in the scientific literature. The method has the potential to accelerate drug and materials discovery by enabling chemists to plan better syntheses faster and by enabling fully automated robot synthesis. The system was tested on a set of 423,731 molecules, and it was able to find synthesis routes for 92% of them within 60 seconds. The method was also tested in double-blind AB tests, where chemists were asked to choose between routes generated by the system and those found in the literature. The chemists showed a significant preference for the routes generated by the system. The method has the potential to revolutionize chemical synthesis planning by providing a fast and accurate way to find synthesis routes.This paper presents a novel approach to planning chemical syntheses using Monte Carlo Tree Search (MCTS) combined with deep neural networks. The method, called 3N-MCTS, uses three neural networks: an expansion policy network, a rollout policy network, and an in-scope filter network. The expansion policy network guides the search by suggesting possible transformations, the rollout policy network predicts the best transformations, and the in-scope filter network ensures that only feasible reactions are considered. The system was trained on 12 million reactions, representing essentially all reactions ever published in organic chemistry. The method outperforms traditional search methods in terms of speed and the number of molecules solved. It can solve almost twice as many molecules and is 30 times faster than traditional methods. The system can generate routes that are indistinguishable from those found in the scientific literature. The method has the potential to accelerate drug and materials discovery by enabling chemists to plan better syntheses faster and by enabling fully automated robot synthesis. The system was tested on a set of 423,731 molecules, and it was able to find synthesis routes for 92% of them within 60 seconds. The method was also tested in double-blind AB tests, where chemists were asked to choose between routes generated by the system and those found in the literature. The chemists showed a significant preference for the routes generated by the system. The method has the potential to revolutionize chemical synthesis planning by providing a fast and accurate way to find synthesis routes.