RaceBench is a benchmarking tool for comparing the JavaScript code generation capabilities of various AI models (LLMs). It takes a set of model responses to a creative coding prompt, extracts runnable JS code, and generates a beautiful, sortable, and searchable HTML dashboard to visually compare results.

Modern LLMs can generate not just text, but also executable code. RaceBench was created to:

• Compare how different models handle complex, creative coding tasks (like games, visualizations, etc.)
• Provide a fair, visual, and interactive way to explore and analyze model outputs
• Help researchers, developers, and enthusiasts see strengths and weaknesses of each model in a real, runnable context

• You provide a race.json file with model responses (see example prompt in the UI)
• Run the get-results.js script (Node.js required)
• The script extracts the last runnable JS block from each response, gathers stats, and generates:
  • index.html: a homepage with a grid and sortable table of all results
  • One HTML file per model, each showing a live preview and raw response
• The UI uses Tailwind CSS and simple-datatables for a modern, interactive experience
• You can toggle between grid and table views, sort by any column, search, and open each result in a new tab

1. Requirements: Node.js (for running the script), a browser (for viewing results)
2. Place your race.json in the project directory
3. Run: node get-results.js
4. Serve the directory with any static file server (e.g. python3 -m http.server)
5. Open index.html in your browser
6. Explore, sort, and compare!

The main script is get-results.js. You can run it with optional arguments to control input and output:

node get-results.js [--input <inputfile>] [--outdir <outputdir>]

• --input <inputfile>
  Path to the input JSON file (default: ./race.json).
  Example:

  node get-results.js --input mydata.json

• --outdir <outputdir>
  Directory where the HTML files will be generated (default: current directory).
  Example:

  node get-results.js --outdir ./public/results

You can combine both options:

node get-results.js --input mydata.json --outdir ./public/results

If you don't specify any options, it will use race.json in the current directory and output files there.

• Beautiful, responsive UI (dark mode supported)
• Toggle between card grid and sortable table
• All columns sortable, with search and pagination
• Each model result opens in a new tab with live preview and raw output
• No backend required: fully static, portable, and easy to share

• UI: Tailwind CSS, simple-datatables
• Created by didi and contributors

MIT