Arena, a platform that originated as a UC Berkeley research project in 2023, closed a $200 million Series B at a $3.1 billion valuation, the company announced Thursday. The round more than doubled its $1.7 billion valuation from January.
The 10-month sprint saw annualized revenue climb from $30 million to $100 million by June. Lightspeed Venture Partners and Khosla Ventures co-led the round, joined by Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Endeavor Catalyst, a16z and Felicis.
Arena's core offering is free: users submit prompts and rate AI model outputs, generating tens of millions of monthly visits. The company monetized in September by launching AI Evaluations, a commercial service selling detailed performance analytics to model labs and enterprises.
The business model exploits a genuine market gap. AI labs increasingly recognize their models are gaming benchmarks rather than performing in real-world conditions. Enterprises, meanwhile, need independent validation of which models work for their specific use cases. Static benchmarks cannot provide that.
Arena addressed this by expanding its leaderboard to include alignment rankings—measuring whether models follow user intent, avoid unauthorized actions, and resist deception. Preliminary results show OpenAI's models at the top, with Claude Opus 5.5 and Claude Fable ranking sixth and ninth.
The competitive moat is structural: Arena aggregates proprietary behavioral data from a massive free user base, creating evaluation insights no benchmark can replicate. As enterprises standardize model selection, independent third-party validation becomes table stakes for adoption. Arena has positioned itself as that validator.
