In April 2023, a group of researchers at UC Berkeley put a simple website online. You typed a question, two anonymous chatbots answered, and you voted for the better reply. Only after you voted did the site reveal which models you had been judging.
Three and a half years later, that voting booth is a $3.1 billion company. On October 8, 2026, Arena, the startup formerly known as LMArena and before that Chatbot Arena, announced a $200 million Series B co-led by Lightspeed Venture Partners and Khosla Ventures. Its valuation has nearly doubled in about ten months.
The Arena AI funding round is one of the clearest signs yet that, in the AI race, the people keeping score can be worth almost as much attention as the players. Here is how it happened, what the numbers say, and what founders can learn from it.
Arena raised $200 million at a $3.1 billion valuation on October 8, 2026, up from $1.7 billion in January, and says it now earns more than $100 million in annualized revenue.
What Arena Announced
In a blog post on its own site, Arena confirmed the $200 million round and the $3.1 billion valuation. Lightspeed and Khosla led. New investors include Salesforce Ventures, 01 Advisors, Dell Technologies Capital, Acrew Capital and Endeavor Catalyst. Existing backers a16z, Felicis, AMP PBC, QuantumLight and The House Fund also put in more money.
The company shared a set of usage numbers alongside the raise. It says it has logged 350 million sessions and 62 million votes across text, vision, code, search, video and image, run more than 1,000 new model evaluations, and draws tens of millions of monthly visitors from more than 150 countries. Its newer Agent Arena, which tests AI agents on real tasks, passed 7 million sessions in under five months.
Arena did not give a detailed plan for the money. It said the funding will support its work on real world AI evaluation and on building its new Alignment Index properly.
How a Berkeley Side Project Became a Unicorn
Arena was founded by Wei-Lin Chiang, Anastasios Angelopoulos and Ion Stoica, the Berkeley professor who also co-founded Databricks. Chatbot Arena launched on April 24, 2023, as an academic experiment in ranking chatbots by human preference instead of by fixed test questions.
The idea caught on fast. As new models arrived almost weekly, the Arena leaderboard became the scoreboard that AI labs quoted in their launch posts. Labs also began testing unreleased models on the site under codenames. Google's Gemini 2.5 Flash Image appeared there as "Nano Banana" before its release, and OpenAI's GPT-5 was tested under the name "summit."
In April 2025, LMArena incorporated as an independent company. A month later it raised a $100 million seed round at a $600 million valuation, with a16z, UC Investments, Lightspeed, Felicis and Kleiner Perkins taking part.
The business model arrived in September 2025, when it launched AI Evaluations, a paid service that gives model labs and large companies detailed performance analytics drawn from community ratings. That product changed everything. By January 6, 2026, when it closed a $150 million Series A at about $1.7 billion led by Felicis and UC Investments, TechCrunch reported its annualized revenue was about $30 million. Three weeks later, LMArena renamed itself simply Arena.
Arena by the Numbers
| Milestone | Date | Figure |
|---|---|---|
| Chatbot Arena launches at UC Berkeley | April 24, 2023 | Free public leaderboard |
| Spins out as a company | April 2025 | LMArena incorporated |
| Seed round | May 2025 | $100M at $600M valuation |
| AI Evaluations launches | September 2025 | First paid product |
| Series A | January 6, 2026 | $150M at about $1.7B |
| Rebrand to Arena | January 28, 2026 | Name change |
| Revenue milestone | June 2026 | $100M+ annualized run rate (company figure) |
| Series B | October 8, 2026 | $200M at $3.1B |
Add it up and Arena has raised about $450 million in roughly 17 months. The revenue numbers are the company's own. Arena is private and has not published audited accounts, so read the $100 million figure as a run rate the company reports, not verified annual revenue.
READ ALSO: Anthropic Raises $65B, Overtakes OpenAI at $965B Valuation
The New Bet: Measuring Whether AI Behaves
The raise came with a new product. Arena launched an Alignment Index that scores frontier models on how they behave as agents, not just how clever their answers are. It starts with three signals: unauthorized action, when a model does things the user never asked for; false attribution, when it credits the user with statements or intentions their own evidence contradicts; and deceptive completion, when it claims a task is finished when it is not.
Arena's pitch is blunt. It says "static benchmarks break down once models recognize they're being tested," and that the world needs "a neutral third party to measure how safe and aligned AI actually is." According to TechCrunch, OpenAI's models hold the top spots on the first, preliminary alignment ranking, with Anthropic's Claude Opus 5.5 in sixth place. Arena itself calls these early results, so treat the ranking as a first snapshot rather than a final verdict.
The Criticism Arena Still Has to Answer
Arena's rise has not been smooth. In April 2025, Meta's Llama 4 Maverick shot near the top of the leaderboard, but the version on Arena was an experimental build that differed from the model the public could download. LMArena updated its policies and said Meta's conduct did not match what it expects from model providers.
Weeks later, researchers from Cohere Labs, Stanford, MIT, Princeton and other institutions published "The Leaderboard Illusion," a paper that has not been peer reviewed. It argued that a few large labs could privately test many model variants and publish only their best scores. It counted 27 private Llama 4 variants tested by Meta before launch, and estimated that Google and OpenAI had each received about a fifth of all Arena data. LMArena rejected the claim that some providers were treated unfairly.
The tension has not gone away. Arena now sells evaluation services to some of the same labs it ranks. Its value depends on staying trusted as a neutral referee, and every new paying customer makes that promise harder to prove.
Why This Matters
Arena's valuation tells you where AI money is flowing in late 2026. The biggest checks still go to model builders, but investors are now paying up for the tools around them: the evaluators, routers and testing platforms that help companies decide which model to trust. With hundreds of models on the market, choosing one has become a business problem worth paying for.
It also shows how fast a free community product can turn into revenue once it becomes a standard. Arena gave its leaderboard away for two years, built the habit, and then sold the deeper data behind it.
READ ALSO: Manus Raises $500M After China Blocked Its $2B Sale to Meta
Lessons for Founders
Founders building anywhere, from San Francisco to London to Lagos, can take three lessons from Arena.
- Become the reference point first. Arena did not charge anyone until labs were already quoting its rankings in their launch posts. Being the default source gave it pricing power later.
- Turn free usage into paid insight. The free leaderboard produced millions of votes. AI Evaluations packaged that data for buyers with budgets, and revenue grew from about $30 million to more than $100 million in roughly six months, by the company's own figures.
- Treat trust as your product. The Llama 4 episode and "The Leaderboard Illusion" showed that one credibility scare can hurt a ranking business more than any rival. Clear, public rules are part of the product.
What to Watch Next
Watch three things. First, how the Alignment Index holds up once labs and outside researchers start testing its method. Second, whether big buyers keep paying for evaluations as labs build stronger testing tools of their own. Third, whether Arena publishes clearer rules on private testing and paid work for model providers, which would go a long way toward answering its critics.
Frequently Asked Questions
How much did Arena raise in October 2026?
Arena raised $200 million in a Series B announced on October 8, 2026. Lightspeed Venture Partners and Khosla Ventures co-led the round, which valued the company at $3.1 billion.
Is Arena the same as LMArena?
Yes. The project started as Chatbot Arena at UC Berkeley in 2023, became the company LMArena in April 2025, and renamed itself Arena on January 28, 2026.
How does Arena make money?
The public leaderboard is free. Arena earns most of its money from AI Evaluations, a paid service launched in September 2025 that sells detailed model performance analytics to AI labs and enterprises. The company says it passed $100 million in annualized revenue in June 2026.
Who founded Arena?
Arena was founded by Wei-Lin Chiang, Anastasios Angelopoulos and Ion Stoica, a UC Berkeley professor and co-founder of Databricks.
