Paradigm Puzzlesexperiments by ScoreBench
Paradigm Puzzles

ScoreBench

Choose your harness, model, and skills. Get a launch prompt for your coding agent.

Challenge details

An account is required to launch runs. Sign up or log in after setup.

Sharing activity
Reproduce an experiment

Import a reproducibility file

Load a scorebench.experiment.recipe JSON file from disk, paste it, or fetch a URL that permits browser access. Review every field before creating fresh runs.

Challenge

Anthropic Take-HomeSubmission: perf_takehome.py
Primary outcomeBest official cyclesLower is better

Run setup

Skills
Add a skill

Prompt

View default prompt

        
        
        
      


      
      
      
Identical settings, independent runs.

      
    


    

      
Challenge level

      

        
Per run