LMArena, now just called Arena, is where a lot of people go to see which AI model is actually best. Instead of benchmarks the companies can tune for, it ranks models by having regular people vote on real answers. The old lmarena.ai address now redirects to arena.ai.
How it works
- You type a prompt and get answers from two anonymous models side by side
- You vote for the one you like better, or call it a tie
- Only after you vote do you see which models you were comparing
- Millions of these votes feed a public leaderboard that ranks models with an Elo-style score, like chess
There are separate leaderboards for text, coding and web development, vision, image generation and search, so you can see which model is strongest at the thing you care about. It’s free to use, and you get to try the newest models, sometimes before they’re officially released.
Where it came from
It started in 2023 as Chatbot Arena, a research project from LMSYS at UC Berkeley, and grew into its own company. It’s not perfect. Some people argue that models get tuned to win votes, but it’s still one of the most watched leaderboards in AI.
Why I saved it: it’s the first place to look when a new model drops and you want to know if it’s actually better or just better marketed.