RushHourEval
~/Benchmarks/RushHourEval
LLM Planning Last Updated: August 2025

RushHourEval: Rush Hour Puzzle Benchmark for LLM Planning

450 text-grounded Rush Hour puzzles across 3x3, 4x4, and 5x5 boards.

The benchmark tests whether language models can produce legal, target-reaching, and optimal move sequences under strict spatial constraints.

3x3 / 4x4 / 5x5 Prompt + Image Viewer Validation Tables
Total puzzles450
3x3150
4x4150
5x5150

Loading benchmark

Puzzle sample

Rush Hour initial puzzle state
Prompt sent to the model 

              

            

          


          

            

              Dataset Viewer
              Loading samples...
            

            

          

        


        

          

            

              
Validated Model Performance

              
Leaderboard Tables

            

            GitHub Repository
          

          

        


        

          

            

              
Validation Plots

              
Performance Visuals