Arena, the crowdsourced leaderboard that ranks AI models by user votes, has raised $200 million in a Series B round at a $3.1 billion valuation, TechCrunch reported Thursday. The price is nearly double the $1.7 billion valuation the company reached in January.

What Happened

Lightspeed Venture Partners and Khosla Ventures led the round. Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Endeavor Catalyst, Andreessen Horowitz and Felicis also took part, along with other investors.

The new money follows rapid revenue growth. Arena's annualized revenue run rate stood at about $30 million at the time of its January Series A and reached $100 million in June, according to TechCrunch.

RelatedProofpoint to Buy Varonis for $6.3 Billion, Citing Risks of AI Agents' Data Access

Arena began in 2023 as a research project at the University of California, Berkeley. Its free site asks users to submit a prompt, compare answers from two anonymous models and pick the better one; those votes feed the public rankings. TechCrunch said the site draws tens of millions of visitors a month.

The company's revenue comes mainly from AI Evaluations, a paid service introduced in September 2025. It gives model developers and enterprises detailed breakdowns of how their models perform, drawn from the community's comparisons.

Why It Matters

Model makers now ship new versions every few weeks, and buyers want a way to compare them that the vendors do not control. Arena casts itself as a neutral third party, and its rankings are routinely cited in model launch announcements. The funding shows investors are willing to pay a premium for that position at a time when traditional benchmarks have lost credibility because models can be tuned to score well on them.

The company is also moving into safety measurement. It has added an alignment leaderboard that tracks three kinds of failure: taking actions a user did not request, attributing information to the wrong source and claiming to have finished a task that was not completed. In the early rankings, OpenAI models lead, while Anthropic's Claude Opus 5.5 and Claude Fable place sixth and ninth.

Context

Arena's influence has also drawn criticism. In 2025, Meta faced complaints after it emerged that a Llama 4 model it had entered on the leaderboard was an experimental version tuned for chat rather than the one released to developers, and outside researchers questioned whether large labs could test many private variants before publishing their best result. Arena later updated its policies on how labs submit models.

Those disputes go to the heart of its business: paying customers include the same labs whose models it ranks. Keeping the public leaderboard trusted while selling evaluation services to those labs is the balance the company must hold as it grows.

What to Watch

Arena plans to widen its paid evaluation work and expand safety-focused rankings such as the alignment leaderboard. Key questions are whether enterprises, not only model labs, become a large share of revenue, and whether the alignment scores gain the same weight in launch announcements that the head-to-head rankings now carry.