The academic research group behind Chatbot Arena, the widely followed platform for ranking artificial intelligence models, has established a formal company, Arena Intelligence Inc., which will operate under the brand name LMArena.
As detailed in an official company blog post on Thursday, the move is intended to secure the necessary funding and structure for significant platform upgrades and accelerated growth.
The venture is being led by key figures from its academic origins, including recent UC Berkeley postdoctoral researchers Anastasios Angelopoulos and Wei-Lin Chiang, alongside UC Berkeley professor and notable tech entrepreneur Ion Stoica, a co-founder of Databricks and Anyscale.
From University Project to Industry Staple
Launched in early 2023 out of UC Berkeley’s Sky Computing Lab by researchers from multiple institutions, Chatbot Arena rapidly gained prominence in the AI sector.
It offered a novel approach to evaluating large language models by crowdsourcing human feedback through blind head-to-head comparisons. Users interact with two anonymous models simultaneously and vote for the preferred response, generating dynamic Elo-based rankings (a method commonly used in chess to rate player skill levels).
This system created a public leaderboard that became influential, quickly attracting a million visitors monthly. Major AI labs like OpenAI, Google, and Anthropic utilize the platform, sometimes even using it as a testing ground for unreleased models. Angelopoulos stated the underlying vision remains constant: “Our vision is that this will remain a place where everybody on the internet can come and try to chat and use AI, compare different providers and so on.”
The project initially operated with support from university grants and donations from organizations including Google’s Kaggle platform, venture capital firm Andreessen Horowitz, and AI infrastructure company Together AI.
Addressing Scale and Scrutiny
The formation of Arena Intelligence Inc. comes as the platform grapples with the operational demands of its popularity. The team stated that “Becoming a company will give us the resources to improve LMArena significantly over what it is today.”
Coinciding with this, a new beta website launched at beta.lmarena.ai, rebuilt, according to the announcement, to enhance speed, improve the mobile experience, and offer clearer voting options based on community input. Future platform upgrades are slated to include user accounts, chat history saving, and the creation of personal leaderboards.
This organizational shift also occurs amid ongoing discussion about the platform’s methodology and potential biases, and existing scrutiny regarding its methods and funding. Past critiques centered on the subjectivity of user votes (influenced by stylistic preferences and varied abilities to detect AI errors), potential demographic skewing of the user base away from the general public, and transparency regarding the full dataset.
Questions about impartiality also arose from its previous funding structure and the different ways commercial API-based models versus static open-source models interact with the platform.
Pledging Neutrality While Expanding Scope
The LMArena team directly addressed the crucial issue of fairness in their announcement, emphasizing a commitment to impartial evaluation: “Our leaderboard will never be biased towards (or against) any provider, and will faithfully reflect our community’s preferences by design. It will be science-driven.” They underscored their belief that capturing genuine, real-world human preferences is key to advancing AI responsibly for all users and developers.
Beyond refining the main chatbot rankings, LMArena plans a significant expansion of its evaluation activities. The roadmap includes bolstering support for open research and launching more specialized testing environments. The company’s website already lists active projects like WebDev Arena, RepoChat Arena, and Search Arena, with plans for future arenas dedicated to vision models, AI agents, and AI red-teaming exercises.
Funding and Future Operations
While Arena Intelligence Inc. intends to raise venture capital to support its growth and platform development, specifics about potential investors or funding targets have not been shared. The company is also still determining its long-term business model. One possibility, mentioned by Stoica to Bloomberg, could involve charging companies for evaluating their models on the platform. Successfully navigating the path to secure funding and potentially introduce revenue streams, while rigorously maintaining the platform’s neutrality and scientific credibility, will be central to LMArena’s transition from its academic origins into a sustainable commercial entity.
