
When using chat models in web-ui, do not share private, sensitive, or personal information. Unlike in real-life situations with professionals (such as doctors or lawyers), there is no guarantee of confidentiality or privacy. Always protect your data.
Arena is a neutral AI model comparison platform, formerly known as LMArena and originally associated with Chatbot Arena. It lets builders, researchers, and curious users compare outputs from third-party AI models, vote for the response they prefer, and explore public leaderboards shaped by community feedback.
Arena measures model performance through real prompts and human preference votes rather than relying only on a fixed test set. The blind format reduces the influence of brand recognition during a battle, while the public leaderboard summarizes how models performed across community comparisons.
The ranking is useful evidence about user preference in Arena's evaluation setting, not a universal measure of model quality. Results can vary with the prompts, voters, model availability, and modality being compared.
In Battle Mode, you submit a prompt and review two responses side by side. After choosing the stronger response, the model identities can be revealed and the vote can contribute to the wider evaluation. You can also use Arena as a chat interface for testing leading models with your own tasks.
The platform extends beyond text chat to image and code models, so its leaderboards cover several kinds of generative AI. Arena accepts file inputs, but prompts are processed by third-party AI providers. Conversations and some personal information may also be disclosed publicly, so sensitive or confidential data should not be submitted.
Arena uses pairwise preference votes as its core feedback signal instead of a conventional star rating. That gives users a direct role in the rankings, but it also means the results reflect the mix of prompts and preferences represented on the platform. Model responses may be inaccurate, and leaderboard positions should be read with that limitation in mind.
Arena currently offers the service free of charge. Its terms reserve the right to charge for the service or individual features in the future.
Arena may share prompts with third-party AI providers or disclose chats publicly. Avoid personal, confidential, or sensitive information.
It is a free service, not a free trial. Usage limits apply, and Arena does not offer a paid plan that removes those limits.
Arena Intelligence, Inc. runs it. The platform was created by researchers from UC Berkeley and was previously called LMArena.
Its rankings reflect human preferences from submitted prompts, so they can inherit voter and task biases. They do not measure universal model quality.
Arena is venture-backed. Its January 2026 Series A was led by Felicis and UC Investments, with several other technology investors.
Ask specific questions about this tool.