Arena, the ranking of AI models, raises $200 million to $3.1 billion and launches an agent alignment index

Valued at $1.7 billion in January, the Berkeley-born platform exceeds $100 million in annualized revenue and now wants to rate agents who act without authorization.

Arena, the valuation platform that emerged from Berkeley's Chatbot Arena project, has announced a $200 million Series B round that values it at $3.1 billion. The round is co-led by Lightspeed Venture Partners and Khosla Ventures, with Salesforce Ventures and Dell Technologies Capital among others. In January, under the name LMArena, the company was valued at $1.7 billion; it has since claimed more than $100 million in annualized revenue, derived from its commercial evaluation offering.

Measuring agent gaps

The lifting is accompanied by the Alignment Index, a new ranking that no longer measures user preferences but the slippages of agents in real conditions: actions taken beyond what was authorized, tasks declared completed when they are not. It covers more than twenty models; OpenAI's GPT-6.1 Sol comes out on top, ahead of Anthropic's Claude Opus 5.5.

Arena justifies this orientation by the limitation of static tests, which the models end up acknowledging, and presents itself as a neutral third party capable of observing the behavior of AIs once in the hands of the public. There remains the criticism already levelled at its historical ranking: a participatory system can be manipulated where its safeguards are weak.