# Arena (LMArena)

> A community-powered platform for evaluating and comparing frontier AI models through real-world human feedback, featuring a public leaderboard and battle mode.

Arena (formerly LMArena) is a community-powered platform created by researchers from UC Berkeley for understanding AI model performance in real-world conditions. The platform lets users interact with frontier AI models, compare their outputs side-by-side in "Battle Mode," and submit feedback that shapes a public leaderboard grounded in actual human preference data. According to the About page, tens of millions of builders, researchers, and creative professionals use Arena to engage with frontier models.

## What It Is

Arena is an AI model evaluation and benchmarking platform that collects human preference signals at scale. Users can prompt multiple AI models simultaneously, vote on which response is better, and contribute to a continuously updated public leaderboard. The platform's core thesis is that real-world human feedback is a more reliable signal of model quality than synthetic benchmarks alone. The project originated from UC Berkeley research and has since grown into a broader community and commercial offering under the "Arena" brand.

## Battle Mode and Core Workflow

The flagship feature is Battle Mode, where two AI models respond to the same prompt and users vote on the better answer. This blind comparison format is designed to reduce bias and generate preference data that reflects genuine user needs. The homepage also highlights practical use cases such as creating landing pages, building dashboards, making browser games, converting designs to code, building full-stack apps, and launching storefronts — suggesting the platform targets developers and builders as a primary audience.

## Leaderboard and Evaluation Services

Arena publishes a public leaderboard ranking AI models based on aggregated human preference votes. The platform also offers an "AI Evaluations" service aimed at enterprises, model labs, and developers, providing comprehensive evaluation grounded in real-world human feedback. This commercial arm is positioned as a way for organizations to benchmark their models against real user preferences rather than static test sets.

## Community and Research Roots

The platform was built by UC Berkeley researchers and maintains a strong community orientation. Arena operates a Discord server, X/Twitter account, LinkedIn page, and YouTube channel to connect researchers, developers, and AI enthusiasts. The About page frames the mission as measuring and advancing the frontier of AI for real-world use, with a vision of building the foundation for everyone to understand, shape, and benefit from AI.

## Rebranding: LMArena to Arena

The platform was previously known as LMArena and has since rebranded to Arena (arena.ai). The lmarena.ai domain now redirects to the new brand. The GitHub organization remains under the lmarena handle, and the blog repository (lmarena.github.io) is open-source under the MIT License. The core platform itself is a web-based service, not an open-source product.

## Features
- Battle Mode for side-by-side AI model comparison
- Public AI model leaderboard based on human preference votes
- Support for frontier AI models
- Real-world human feedback collection
- AI Evaluations service for enterprises and model labs
- Practical use-case prompts (landing pages, dashboards, games, apps)
- Community-driven model ranking
- Search conversation history

## Integrations
Multiple frontier AI model providers

## Platforms
WEB, API

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://lmarena.ai
- Documentation: https://arena.ai/faq
- Repository: https://github.com/lmarena/lmarena.github.io
- EveryDev.ai: https://www.everydev.ai/tools/lm-arena