The Site That Ranks AI Chatbots Just Raised $200 Million
Arena, the website that ranks AI chatbots, has raised $200 million at a $3.1 billion valuation.
The funding round was led by Lightspeed Venture Partners and Khosla Ventures. Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Endeavor Catalyst, a16z and Felicis also joined in, according to TechCrunch. Arena announced the funding on Oct. 8, along with a new AI Alignment Index.
The new funding nearly doubles its value from January 2026. At that time, Arena raised $150 million at a $1.7 billion valuation. That earlier round had tripled its value in about eight months.
Arena began in 2023 as a research project at UC Berkeley. The idea was simple, like a blind taste test. Show two AI answers without names, ask people which is better, and use the votes to rank models. The site now says it gets tens of millions of visitors each month.
That traffic became a business in September 2025. Arena started selling AI Evaluations to AI labs and large companies. Instead of only a free public ranking, it now offers paid tests, repeat checks that catch problems when models are updated, and help for companies choosing which model to buy.
Sales grew fast. Arena said yearly revenue on its current pace was $30 million at the time of its earlier funding. That pace had already passed $30 million in December, less than four months after the paid product launched. By June 2026, the company said it had reached $100 million on that same yearly-pace measure.
The new release adds safety testing. Arena added an alignment ranking that grades models on three problems. Acting without permission, making up sources, and giving misleading answers that look correct. It described the AI Alignment Index, described in Arena, as a way to test AI as people actually use it.
The broader context here is a change in how AI is tested. Public votes help people get a rough idea of which model is better. They do not tell a company if a model is safe to use at work. Test questions can leak, small wording changes can shift results, and a general score says little about a specific job. Paid tests use private questions and track results over time. They ask a practical question. Will a new version break things that already work.
In my view, this safety ranking is worth watching. Acting without permission, making up sources, and misleading answers do not show up in simple speed or accuracy scores. They show up when AI uses tools, cites documents, or handles multi-step tasks. A public comparison could change how companies test AI before buying it. It will only help if Arena shows its methods clearly.
The hopeful possibility here is steady testing instead of one-time grades. Arena has large public traffic on one side and business customers on the other. If it works, choosing a model will depend less on last month's winner and more on what holds up in daily work.
