The AI industry's favorite testing playground is growing up. Chatbot Arena, the crowdsourced benchmarking platform that has become a crucial testing ground for AI models, announced today it's forming a company called Arena Intelligence Inc., operating under the brand name LMArena.
Key Points:
- Rebranding as LMArena with plans to maintain neutrality while expanding capabilities
- Currently attracts one million visitors monthly who rank AI models head-to-head
- Led by UC Berkeley researchers who are now transitioning to company roles
What began in early 2023 as a scrappy research project from UC Berkeley's Sky Computing Lab has evolved into a significant force in AI evaluation, attracting a million visitors monthly who compare models in head-to-head competitions. The platform's leaderboards have become closely watched signals of model quality across the industry.
The new company will be led by the project’s original team: Anastasios Angelopoulos and Wei-Lin Chiang—both recently postdocs at Berkeley—and their advisor Ion Stoica, a heavyweight in cloud computing and co-founder of Databricks and Anyscale. While exact titles are still in flux, the goal is clear: scale the platform, fix longstanding usability bugs, and build new features based on community feedback.
The platform has carved out a unique position in the AI ecosystem by providing neutral, user-driven assessments of model capabilities. Many major AI developers, including OpenAI, have used Chatbot Arena to test new models before wider releases. This neutrality appears central to the team's vision for the company's future.
"LMArena will be staying true to its original mission. It will remain a neutral, open platform for testing and evaluating AI models," the team wrote in their announcement. "Our leaderboard will never be biased towards (or against) any provider, and will faithfully reflect our community's preferences by design."
To that end, the team also launched a beta version of the site at beta.lmarena.ai, a rebuilt platform that improves speed, mobile experience, and voting clarity—common complaints from long-time users. Features like logins, chat history, and personalized leaderboards are coming soon, along with new experimental spaces like WebDev Arena and RepoChat Arena.
The company hasn’t finalized a business model, though one option being explored is charging providers for model evaluations. Stoica confirmed they intend to raise money to fund the growth, but declined to share fundraising details.
For everyday AI users and AI-curious professionals, the platform offers a rare opportunity to directly compare leading systems without marketing spin – voting with their clicks on which models actually perform best in practical scenarios.
As funding flows into the new company, the question remains whether Arena Intelligence can maintain its academic neutrality while developing a sustainable business. For now, the team is emphasizing transparency and community trust as core values, recognizing that its credibility is its most valuable asset.