What is LM Arena?
LM Arena (formerly Chatbot Arena) is a public, web-based evaluation platform that pits two anonymous models against each other on user-supplied prompts. After reading both responses, the user picks the better one; the system then reveals the identities and updates each model’s Elo rating.
Highlights
- 3 M+ human votes powering a robust Elo ladder for GPT-4o, Gemini 2.5-Pro, Claude Opus 4 and open-source models.
- Live leaderboard plus task-specific tables (e.g., MMLU, Arena-Hard-Auto).
- Open API & research data – frequently cited in papers studying evaluation bias and prompt sensitivity.
Use cases
Benchmarking, model-routing, marketing, and community engagement for new LLM releases.
