JavaScript Coding

Evaluation of AI's ability to create interactive browser-based games using JavaScript

Evaluation Type: MODEL

About This Skill

This skill evaluates model short-horizon coding performance in creating complete, functional JavaScript games that render in HTML5 Canvas with real-time interaction, animations, and game logic. Models are judged on code quality, visual presentation, and successfully delivering the requirements.

Total Models: 51
Top Score: 1671
Average Score: 1502

Performance Comparison

1. Google: Gemini 2.5 Pro
   Score: 1671±65.3
   Link

2. OpenAI: GPT-5 Mini
   Score: 1648±67.2
   Link

3. OpenAI: GPT-5
   Score: 1643±77.2
   Link

4. Anthropic: Claude 3.7 Sonnet (thinking)
   Score: 1607±45.1
   Link

5. Anthropic: Claude Opus 4
   Score: 1586±41.8
   Link

6. Anthropic: Claude Opus 4.1
   Score: 1578±44.0
   Link

7. Anthropic: Claude Sonnet 4
   Score: 1574±40.8
   Link

8. Qwen: Qwen3 235B A22B Instruct 2507
   Score: 1574±39.2
   Link

9. Google: Gemini 2.5 Flash
   Score: 1564±42.0
   Link

10. **OpenAI:** GPT-4.1 Mini  
    **Score:** 1561±36.6  
    [Link](https://openrouter.ai/models/openai/gpt-4.1-mini)
    

... (more data omitted for brevity)

Evaluation Methodology

Tests models' ability to create complete, functional JavaScript games with advanced features using simple canvas rendering.

Skill-Specific Setup

• Evaluation Type: Single-prompt technical implementation
• Evaluation Count: 7 distinct game challenges (Conway's Game of Life, Flight Simulator, Maze Generator, Procedural FPS Map, Racing Game, Sorting Visualizer, Space Invaders)
• Task Complexity: Each evaluation requires implementing a complete interactive game with multiple features, UI controls, and smooth performance.
• Expected Output: Valid JavaScript code only, no explanatory text or markdown formatting.

Judge Exposure

What Judges Saw: Two complete JavaScript implementations side-by-side

• Model A's complete code solution
• Model B's complete code solution
• Judge Task: Evaluate functionality, correctness, and feature completeness without penalizing different coding styles.

Tournament Structure

Swiss Round System: Each model competed against others across all 7 game development challenges. Judges compared pairs of implementations, selecting the superior solution based on technical merit, feature completeness, and code quality. Rankings were determined using Bradley-Terry scoring across all comparisons.