Showdown

Which AI model is actually the best? We aggregate 20+ benchmarks so you don't have to.

What is Showdown?

Tired of cherry-picked benchmarks and marketing hype? Showdown provides transparent, community-maintained rankings of AI language models across real-world categories:

• Coding - Can it actually write working code?
• Reasoning - PhD-level science, complex logic
• Agents & Tools - Function calling, browser automation
• Math - From algebra to competition problems
• Multimodal - Vision understanding
• Multilingual - Beyond English
• Conversation - Creative writing, instruction following

All data is open. All methodology is transparent. All contributions are welcome.

Quick Start

Visit showdown.best to explore the rankings.

Want to run it locally?

git clone https://github.com/verseles/showdown.git
cd showdown
npm install
npm run dev

How Rankings Work

We aggregate scores from 20+ industry benchmarks, weighted by practical importance:

Scoring:

• Percentage benchmarks used directly
• Elo scores normalized to 0-100
• Missing data? We estimate using smart imputation (marked with * in UI)
• Final score = weighted average across categories

Imputation Methods

When benchmark data is missing, we use two estimation methods:

1. Superior Model Imputation (green *): For "thinking" variants, we calculate their expected superiority over the base model using benchmarks where both have real data, then apply that ratio to missing benchmarks. More reliable since it's based on real performance differences.

2. Category Average (yellow *): Falls back to averaging other benchmarks in the same category. Less reliable but ensures all models can be compared.

Note: Estimated values are clearly marked and should be replaced with real data when available. See UPDATE.md for details.

Contributing

Found an outdated score?

Open an issue with the correct value and source.

Want to add a model?

Open an issue with available benchmark scores.

Ready to submit a PR?

1. Fork this repo
2. Edit data/showdown.json
3. Run ./precommit.sh to validate your changes
4. Submit PR - our CI validates the data automatically
5. Get merged!

Tech Stack

• Frontend: Svelte 5 + SvelteKit (static site generation)
• Data: Single JSON file - easy to edit, easy to validate
• Hosting: Cloudflare Pages - fast worldwide
• CI/CD: GitHub Actions - automated validation on every PR

Data Sources

Rankings aggregate data from trusted sources:

• SWE-Bench - Real GitHub issue resolution
• GPQA - PhD-level questions
• BFCL - Function calling
• LMArena - Human preferences
• Artificial Analysis - Speed metrics

License

AGPL-3.0 - Keep it open!

Built with Svelte. Hosted on Cloudflare. Made for the community.