Arena

Arena, previously known as LMArena and Chatbot Arena, is a public web platform where people pit anonymous AI chatbots against each other. You type in a prompt, two unnamed models respond, and you vote on which answer you prefer. After you cast your vote, the identities of the models are revealed. There is also a "Direct" mode where you can pick a specific model to test on your own.

The platform has become a popular testing ground for AI companies. OpenAI, Google DeepMind, and Anthropic have all submitted their models for evaluation. It is also a place where companies try out new models before their official release. For instance, Chinese startup DeepSeek quietly tested prototype models on Arena months before its R1 model made headlines in Western media. OpenAI's GPT-5 appeared under the codename "summit," and Google DeepMind's Gemini 2.5 Flash Image, an image generation and editing model, was tested as "Nano Banana."

Arena's crowd-sourced rankings have made it a go-to reference for AI enthusiasts and researchers, but the methodology is not without flaws. Studies have pointed out limitations in how the votes are collected and interpreted, which can skew results. Still, it remains a lively, hands-on way to see how AI models stack up in real-world conversations. [edit]