gr8monk3ys/model-arena

AI Model Arena

Compare AI models head-to-head! Test the same prompt across different models and vote for the best response.

Features

6 Top Open Models

• —Mistral-7B - Fast and efficient, great at reasoning and code
• —Llama-3.1-8B - Meta's latest with strong general capabilities
• —Qwen2.5-7B - Excellent at multilingual tasks, math, and coding
• —Phi-3-mini - Microsoft's compact powerhouse
• —Gemma-2-9B - Google's quality-focused instruction model
• —Zephyr-7B - Aligned for helpfulness and safety

Battle System

• —Run any two models against each other
• —See response times for each model
• —Vote for the better response
• —Track wins on the leaderboard

5 Test Categories

• —Creative Writing - Poetry, stories, creative prompts
• —Coding - Programming challenges and algorithms
• —Reasoning - Logic puzzles and math problems
• —Knowledge - Explanations and factual queries
• —Summarization - Condensing complex topics

How to Use

1. 1.Enter a prompt or use an example from a category
2. 2.Select two models to compare
3. 3.Click "Start Battle" to generate responses
4. 4.Read both responses and compare quality, accuracy, and style
5. 5.Vote for the better response
6. 6.Check the leaderboard to see which models are winning!

Example Battles

Why This Matters

Different models have different strengths:

• —Some are faster, some more accurate
• —Some excel at code, others at creative tasks
• —Testing helps you choose the right model for your needs

Technical Details

• —All models accessed via HuggingFace Inference API
• —Response times measured for comparison
• —Leaderboard persists during Space session

Configuration

This Space calls the HuggingFace Inference API to run every model. It requires an HF_TOKEN secret — a token with inference access — set in Space Settings → Secrets. Without it, battles will fail.

License

MIT

Author

Built by Lorenzo Scaturchio