Arena

Where AI meets the real world. Formerly LMArena. We measure and advance the frontier of AI through community-driven evaluation. We’re hiring → https://t.co/XBZCrseaWF

Arena is a crowdsourced evaluation platform for AI models, used by builders and labs to measure how systems actually perform when real people put them to work. Originally launched as Chatbot Arena by researchers at UC Berkeley, the platform allows users to vote on which model response they prefer , and that vote data has become a reference point for ranking frontier models. The launch marks the company hitting a $100M annual revenue run rate roughly eight months after introducing its commercial AI Evaluations product, which lets enterprises run their own use cases through the same human-judged comparison system that powers the public leaderboard. The company was co-founded by Anastasios Angelopoulos and Wei-Lin Chiang , with UC Berkeley professor and Databricks co-founder Ion Stoica advising the project before it spun out. LMArena rebranded as Arena as it expanded past pure leaderboard work into evaluation infrastructure that customers can point at their own agents and applications. According to the launch video, the underlying community has cast tens of millions of votes across hundreds of millions of conversations, and that signal is what the commercial product resells as alignment-to-real-use measurement. This launch matters now because evaluation is becoming the bottleneck as AI shifts from single-turn chatbots to agents that complete longer tasks. Arena is pushing into that gap with Agent Arena, a newer product line aimed at scoring task completion, hallucinations, and steerability rather than just pairwise preference. For founders shipping agentic products and investors trying to compare them, Arena is positioning its human-in-the-loop dataset as the benchmark layer the rest of the stack tests against.

Tags

Comments (7)