Pomegra
Arena.ai raised a $200M Series B at a $3.1B valuation, co-led by Lightspeed and Khosla, as its AI model evaluation platform tops $100M in annualized revenue.
Key Takeaways
- Arena.ai closed a $200M Series B at a $3.1B valuation, announced October 8, 2026.
- Lightspeed Venture Partners and Khosla Ventures co-led; valuation is up from $1.7B in January.
- Annualized revenue exceeded $100M, versus about $30M in January.
Lead
Arena.ai, the US company behind the public leaderboard that ranks frontier AI models, raised $200M in a Series B round announced on October 8, 2026. The round values the company at $3.1B. Lightspeed Venture Partners and Khosla Ventures co-led.
Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Acrew Capital and Endeavor Catalyst joined the round. Existing backers a16z, Felicis, AMP PBC, QuantumLight and The House Fund also participated.
What Did Arena.ai Raise, and How Does It Compare With Its Last Round?
The $200M Series B lifts Arena's valuation by roughly 82% in about ten months. In January 2026 the company raised a $150M Series A at a $1.7B post-money valuation. The new round is larger by $50M and priced $1.4B higher.
Revenue explains most of the jump. Arena reported about $30M in annualized revenue in January and passed $100M in annualized run-rate by June. That is a more than threefold increase in five months, which is the kind of curve that supports a doubling of price. The company has not disclosed profitability, burn or the ownership stake sold in this round.
What Does Arena Actually Do?
Arena runs a crowdsourced platform where users compare anonymous model outputs and vote on which is better. Those votes feed public rankings across text, vision, code, search, video, image generation, document analysis and web development. The company says it tracks more than 1,000 models, open and proprietary, and has collected 62 million votes.
The project began in 2023 as a UC Berkeley research effort. Tens of millions of people from more than 150 countries now visit each month, and the platform has logged 350 million sessions in total. It has open-sourced 375,000 data points.
Revenue comes from a different product. Arena launched a commercial AI Evaluations service in September 2025, which lets model developers and enterprises test systems against its voter base and methodology. The free leaderboard drives traffic and credibility. The paid service converts that credibility into contracts.
Why Is Agent Evaluation the New Focus?
Static benchmarks lose value once models are trained against them, and agents are harder to score than chat responses. Arena's chief executive, Anastasios Angelopoulos, has framed the problem as AI advancing faster than the ability to evaluate it, with fixed tests failing once models recognize they are being tested.
The company's Agent Arena product has logged 7 million sessions in less than five months since launch. With the round, Arena also introduced the Arena Alignment Index, built from real-world agent traces. It scores models on three behaviors: Unauthorized Action, where a model goes beyond what the user asked; False Attribution, where it attributes false statements to users; and Deceptive Completion, where it claims a task is finished when it is not. The first release covers more than 20 frontier models.
Does Neutrality Survive the Cap Table?
The central question for Arena is whether a referee can stay credible while taking money from parts of the industry it ranks. Its revenue depends on AI developers and enterprise buyers, and its investor list now includes Salesforce, Dell Technologies and a16z, all of which have exposure to the model market. Arena's defense rests on transparency of method and open data, but the structure invites scrutiny each time a leaderboard position moves.
Competitors are also a factor. Model labs publish their own benchmark results, and enterprises increasingly build private evaluation suites. Arena's advantage is a large pool of human preference data that is expensive to replicate. That advantage holds only while voters keep showing up and the rankings are seen as unbiased.
Outlook
Arena has gone from a university project to a $3.1B company in under three years, with revenue growth that justifies the price better than most rounds of this size. The next tests are whether agent and alignment evaluation turns into a second revenue line and whether enterprise customers pay for third-party scoring at scale. Credibility with the labs being measured will decide both.
