Arena

Arena is the official AI ranking and LLM leaderboard platform, formerly known as LMArena and Chatbot Arena, now operating at arena.ai. The core experience is a blind side-by-side comparison: users submit a prompt and receive responses from two anonymous models, then vote for the better answer. Each vote contributes to the public leaderboard, which ranks large language models, image models, and code models based on real-world human evaluation rather than synthetic benchmarks. The platform hosts a rotating lineup of the latest frontier models, open-weight models, and specialized systems, letting users chat with models directly from the leaderboard page. Beyond text chat, Arena supports evaluation of image models and code models, expanding the classic chatbot arena format into multimodal comparison. The service is free for users, and the project has become one of the most cited community benchmarks in AI research, with researchers and developers referencing its Elo-style rankings when discussing model quality. Users can browse the leaderboard by category, try trending models, and see detailed statistics on model performance. The platform emphasizes transparency: conversations and votes help train and inform the community, and users are warned that inputs are processed by third-party AI providers and responses may be inaccurate. Arena also functions as a community hub where new model releases are quickly added for testing, making it a go-to destination for anyone tracking the state of AI. For researchers, the accumulated preference data is a valuable resource for studying model capabilities and alignment. The voting interface is intentionally minimal, which keeps the focus on model quality rather than platform features. Users can engage in casual conversation, stress-test models with complex prompts, or focus on specific domains such as mathematics, programming, or creative writing. Image and code model battles follow the same anonymous comparison format, letting users evaluate visual outputs and code quality side by side. The public leaderboard pages present model rankings with usage statistics, and new models are continuously added for community testing as they launch. For many developers, Arena has become the first place to check how a newly released model performs in practice, and its community-driven rankings are frequently referenced in industry discussions and academic work.