Hugging Face leaderboards: how to read AI model rankings
A one-minute guide to the online scoreboards that compare AI models, and how to read them wisely.
Have you ever wondered how people decide which AI model is “best”? Often, they check a leaderboard, and many of the most popular ones live on Hugging Face.
What is Hugging Face?
Hugging Face is a website where people share AI models, datasets and demos, a bit like a friendly library for AI. Anyone can browse it, try models in the browser, and see how they compare.
What does a leaderboard show?
A leaderboard is a scoreboard. Models take the same set of tests, called benchmarks, and get ranked by their scores. Some leaderboards test general knowledge or reasoning; others focus on maths, coding or a particular language. One of the best known was the Open LLM Leaderboard, which compared thousands of openly shared language models side by side.
How to read one wisely
- Check what’s being tested. A top score in maths says little about writing a friendly email.
- Look at the size of the gap. Models a fraction of a point apart are usually equally good.
- Try it yourself. The best model is the one that works well for your task.
Curious? Search for “leaderboard” on Hugging Face Spaces and see how many different scoreboards there are.
Want to go deeper? Our guide to testing chatbots fairly explains why scores aren’t the whole story.