The paper introduces the System-1.x Planner, a novel framework that combines fast and slow planning modes using language models (LLMs) to solve long-horizon planning problems. The System-1.x Planner is designed to balance between two planning modes: System-1, which generates plans without explicit search, and System-2, which plans step-by-step through explicit action search. The key contributions of the System-1.x Planner include: 1. **Hybrid Planning**: The planner generates hybrid plans that alternate between System-1 and System-2 sub-plans based on the difficulty of the sub-goal. 2. **Controllability**: A user-specified hybridization factor \( x \) controls the balance between System-1 and System-2 planning, allowing for fine-tuning of performance and computational efficiency. 3. **Flexibility**: The planner can be adapted to different search algorithms (e.g., BFS, DFS, A*) and can switch between neural and symbolic planning modes. 4. **Generalizability**: The planner demonstrates robustness to different search algorithms and outperforms both System-1 and System-2 planners in various planning tasks. The System-1.x Planner is evaluated on two tasks: Maze Navigation and Blocksworld. Results show that the System-1.x Planner outperforms the System-1 Planner, the System-2 Planner trained to approximate A* search, and a symbolic planner (A* search) across different exploration budgets. The planner's ability to balance System-1 and System-2 planning, along with its controllability and flexibility, makes it a promising approach for long-horizon planning tasks.The paper introduces the System-1.x Planner, a novel framework that combines fast and slow planning modes using language models (LLMs) to solve long-horizon planning problems. The System-1.x Planner is designed to balance between two planning modes: System-1, which generates plans without explicit search, and System-2, which plans step-by-step through explicit action search. The key contributions of the System-1.x Planner include: 1. **Hybrid Planning**: The planner generates hybrid plans that alternate between System-1 and System-2 sub-plans based on the difficulty of the sub-goal. 2. **Controllability**: A user-specified hybridization factor \( x \) controls the balance between System-1 and System-2 planning, allowing for fine-tuning of performance and computational efficiency. 3. **Flexibility**: The planner can be adapted to different search algorithms (e.g., BFS, DFS, A*) and can switch between neural and symbolic planning modes. 4. **Generalizability**: The planner demonstrates robustness to different search algorithms and outperforms both System-1 and System-2 planners in various planning tasks. The System-1.x Planner is evaluated on two tasks: Maze Navigation and Blocksworld. Results show that the System-1.x Planner outperforms the System-1 Planner, the System-2 Planner trained to approximate A* search, and a symbolic planner (A* search) across different exploration budgets. The planner's ability to balance System-1 and System-2 planning, along with its controllability and flexibility, makes it a promising approach for long-horizon planning tasks.