Arena, the organization behind the industry-standard Chatbot Arena, has officially transitioned its video model evaluation platform from a restricted Discord environment to a public web interface. This shift marks a significant expansion in how the AI community measures the performance of video generation tools. By opening the "Video Arena" to a global audience, the team aims to collect a more diverse range of prompts and languages, ensuring their leaderboards reflect real-world utility rather than niche experimental use cases.

The core of the experience is "Battle Mode," a blind-test (an evaluation where model identities are hidden to prevent bias) methodology where users compare two anonymous generations side-by-side. Whether starting from a text prompt or an image-to-video request, participants vote on which output is superior. This crowd-sourced data feeds into a dynamic leaderboard featuring heavyweights such as Google's Veo 3 and OpenAI's Sora 2, alongside rising stars like Kling-2.6-pro. To ensure scientific rigor, the platform utilizes confidence intervals to provide statistical certainty for every model's score and ranking.

This launch follows a substantial $150M Series A funding round, signaling Arena's intent to become the definitive third-party arbiter for every frontier model—the most advanced AI systems currently in existence. By allowing users to download and share these generations, Arena creates a flywheel of community engagement that helps quantify visual quality through collective human preference. As participation scales, the platform will continue to refresh its pipeline to capture the evolving landscape of video synthesis.

Arena, the organization behind the industry-standard Chatbot Arena, has officially transitioned its video model evaluation platform from a restricted Discord environment to a public web interface. This shift marks a significant expansion in how the AI community measures the performance of video generation tools. By opening the "Video Arena" to a global audience, the team aims to collect a more diverse range of prompts and languages, ensuring their leaderboards reflect real-world utility rather than niche experimental use cases.

The core of the experience is "Battle Mode," a blind-test (an evaluation where model identities are hidden to prevent bias) methodology where users compare two anonymous generations side-by-side. Whether starting from a text prompt or an image-to-video request, participants vote on which output is superior. This crowd-sourced data feeds into a dynamic leaderboard featuring heavyweights such as Google's Veo 3 and OpenAI's Sora 2, alongside rising stars like Kling-2.6-pro. To ensure scientific rigor, the platform utilizes confidence intervals to provide statistical certainty for every model's score and ranking.

This launch follows a substantial $150M Series A funding round, signaling Arena's intent to become the definitive third-party arbiter for every frontier model—the most advanced AI systems currently in existence. By allowing users to download and share these generations, Arena creates a flywheel of community engagement that helps quantify visual quality through collective human preference. As participation scales, the platform will continue to refresh its pipeline to capture the evolving landscape of video synthesis.