LMArena

What is LMArena?

LMArena is a community-driven evaluation platform for large language models (LLMs) that originated from the Chatbot Arena project by LMSYS Org at UC Berkeley. Rather than relying solely on static technical benchmarks, the platform enables users to submit prompts and compare anonymous responses from two different AI models, voting on which they prefer. The platform has expanded beyond text to include specialized arenas for image generation, web development, coding assistance, and search-augmented models. All votes are aggregated using an Elo ranking system to create a transparent, real-world performance leaderboard that reflects how models actually perform in practical applications. LMArena serves AI model developers, researchers, and everyday users seeking fair, unbiased comparisons of AI systems without the influence of marketing or promotional claims.

How to use LMArena?

To use LMArena, visit the platform and select your preferred arena (Text, Image, Code, or Search). In Battle Mode, enter a prompt and receive two anonymous AI responses; vote for the one you prefer. Your vote updates model rankings on the leaderboard in real-time. Alternatively, use Side-by-Side Mode to directly compare specific models of your choice, or Direct Chat Mode to interact with individual models. No account is required to participate in battles, though some models may require authentication for full access. Over time, your votes contribute to the community-driven leaderboard that reflects true user preferences rather than marketing claims.

LMArena's Core Features

• Blind side-by-side comparison of two AI models with responses hidden until voting is complete.

• Elo-based ranking system that updates model scores dynamically based on user votes.

• Multiple specialized arenas for text, image, web development, coding, and search model evaluation.

• Community-driven leaderboard reflecting real-world human preferences rather than static benchmarks.

• Support for anonymous model comparison to eliminate bias and preserve evaluation integrity.

• No-login access to basic battles, making the platform immediately accessible to all users.

• Real-time ranking updates as users submit votes, showing live competitive standings.

• Comprehensive data collection with open-source human preference datasets available to researchers.

• Multi-turn conversation support and extended context handling for nuanced model evaluation.

• Multilingual and domain-specific testing across diverse real-world scenarios.

• Transparency in model performance based on practical applications rather than promotional claims.

LMArena's Use Cases

• #1

  Compare chatbot capabilities to find the best model for answering questions, writing, or reasoning tasks

• #2

  Evaluate image generation models to discover which produces the highest-quality visuals for your needs

• #3

  Test AI coding assistants to identify the most helpful model for code completion and debugging

• #4

  Benchmark search-augmented language models for research, current events, and information retrieval tasks

• #5

  Discover emerging AI models and upcoming capabilities through real-world community testing

• #6

  Provide feedback on AI model performance to help developers improve their systems before public release

• #7

  Make informed decisions about which AI services to adopt based on transparent, real-world performance data

• #8

  Track progress in the AI field by monitoring how model capabilities evolve over time