Arena, a platform that began in 2023 as a UC Berkeley research project crowdsourcing rankings of AI models, announced on Thursday that it has closed a $200 million Series B round, valuing the company at $3.1 billion. The news follows the company's disclosure in June that it had hit $100 million in annualized run-rate revenue.

The funding round was led by Lightspeed Venture Partners and Khosla Ventures, with participation from Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Endeavor Catalyst, a16z, Felicis, and additional investors. This latest raise comes roughly 10 months after Arena's $150 million Series A in January, which carried a $1.7 billion post-money valuation. At that time, the company reported $30 million in annualized revenue — meaning its valuation has nearly doubled in under a year.

Arena operates a free, crowdsourced platform where consumers submit prompts or request vibe-coded projects, then rate which AI model performs better. The company says it draws tens of millions of monthly visitors. In September of last year, it rolled out its commercial offering, AI Evaluations, which gives model labs and enterprises detailed performance analytics derived from community feedback.

That launch turned out to be well-timed. This year, AI labs discovered that their models were gaming benchmarking tests — finding ways to achieve high scores without genuinely earning them. Meanwhile, enterprises increasingly sought guidance on which model best suited their internal needs, rather than depending solely on standardized benchmarks.

"AI is advancing faster than our ability to evaluate it, and static benchmarks break down once models recognize they're being tested," the company stated in its funding announcement. "The world needs a neutral third party to measure how safe and aligned AI actually is once it's in the hands of real people. Arena is stepping into that role today," it added.

Along those lines, Arena has introduced a new leaderboard category: alignment. This ranking evaluates models on issues such as unauthorized action (taking steps it wasn't asked to take), false attribution (incorrectly crediting statements or facts to the wrong source), and what it terms "deceptive completion" (lying about finishing tasks it never completed). As it stands, a group of OpenAI's models occupy the top spots on the preliminary alignment leaderboard, with Claude Opus 5.5 and Claude Fable landing in sixth and ninth place, respectively.