LMArena
FreeHuman-preference Elo rankings for text, image, and video.
FreeFree tier
Inputs: text, image, video
About LMArena
Created by researchers from UC Berkeley, Arena (formerly LMArena) is a community-powered platform for understanding AI performance in the real world. Tens of millions of builders, researchers, and creative professionals come to Arena to use frontier models and give feedback on their responses, shaping a public leaderboard grounded in real-world use. The platform provides human-preference Elo rankings across text, image, vision, web development, document, image edit, search, and video categories, with confidence intervals and real-time updates.
Key Features
Crowdsourced human feedback evaluations from millions of users
Elo-based ranking system with uncertainty margins
Multiple evaluation categories: text, vision, webdev, document, image, text-to-image, image edit, image-to-webdev, search, text-to-video
Real-time leaderboard tracking frontier AI models
Community-powered platform enabling transparent AI performance comparisons
Pros & Cons
Pros
- Community-driven and transparent evaluation process
- Covers a wide range of AI capabilities (text, image, video, etc.)
- Up-to-date rankings reflecting the latest model releases
- Open and accessible to anyone for free
Cons
- Rankings may reflect popularity bias due to self-selecting user base
- Relies on subjective human preferences rather than objective metrics
Best For
Comparing performance of frontier AI models across diverse tasksBenchmarking models for research and developmentSelecting the best AI model for specific real-world applicationsEvaluating model capabilities in text, vision, image generation, and more
FAQ
What is Arena?
Arena (formerly LMArena) is a community-powered platform created by UC Berkeley researchers that ranks AI models based on real-world human feedback.
How are models ranked?
Models are ranked using Elo ratings calculated from pairwise comparisons made by users. The leaderboard shows scores with confidence intervals reflecting statistical uncertainty.
What categories does the leaderboard cover?
The leaderboard covers text, vision, web development, document, image, text-to-image, image edit, image-to-webdev, search, and text-to-video categories.