Arena Raises $200 Million at $3.1 Billion, and Its Announcement Never Says How It Earns Revenue

Arena, the crowdsourced AI leaderboard formerly called LMArena, raised a $200 million Series B at a $3.1 billion valuation on Oct. 8, according to TechCrunch. Lightspeed Venture Partners and Khosla Ventures led, up from a $1.7 billion post-money valuation at its $150 million Series A in January.

Arena's own Series B post, credited to the Arena Team rather than a named executive, adds Acrew Capital and names Salesforce Ventures, 01 Advisors, Dell Technologies Capital and Endeavor Catalyst as participants. Existing backers a16z, Felicis and AMP PBC returned. The announcement is a company post, and no filing has confirmed the closing.

The multiple fell from 57 to 31

Arena said annualized revenue was $30 million at the Series A and reached $100 million in June, per TechCrunch. Dividing each valuation by the run-rate nearest to it gives roughly 57 times in January and 31 times now, though the Series B price is measured against a figure four months old.

	Series A (Jan. 2026)	Series B (Oct. 2026)
Raised	$150 million	$200 million
Valuation	$1.7 billion post-money	$3.1 billion
Annualized revenue	$30 million	$100 million (June)
Implied multiple	57x	31x

1. January 202630 $ million
2. June 2026100 $ million

Source: TechCrunch, Oct. 8, 2026 (January figure at Series A; June figure as run-rate)

The Series B post states that revenue has passed $100 million. It does not say how Arena earns it. TechCrunch fills the gap: the leaderboard is free, and a product called AI Evaluations, launched in September 2025, sells performance analytics built from community votes to model labs and enterprises.

What the votes are worth

The company says its platform has logged 350 million sessions and 62 million votes across text, vision, code, search, video and image, with tens of millions of monthly visitors from more than 150 countries. Agent Arena logged 7 million sessions in under five months. Those are Arena's counts, and a vote measures preference rather than correctness.

Arena started in 2023 as a UC Berkeley research project. Its post says it "started as a research experiment."

The Alignment Index ranks the labs that pay

Arena launched an Alignment Index the same day. It scores models on unauthorized action, false attribution and "deceptive completion," meaning a claim that a task was finished when it was not. TechCrunch reported a preliminary board on which OpenAI models hold the top spots, Claude Opus 5.5 ranks sixth and Claude Fable ninth. A related Arena research post puts the index at 27 models and about 90,000 agent sessions; the full page text was not retrievable, so its scores and method are unchecked.

The structure deserves a plain statement. The ranked models belong to labs, and labs are among the buyers of AI Evaluations. Neither TechCrunch nor Arena's post describes how, if at all, the two are separated. Arena argues that static benchmarks go stale once models detect a test, and benchmark scores are the subject of Cloudflare's Clef beating TypeSafe's Jev on three tests and losing on one.

The same week, TypeSafe AI announced $870 million at $7.5 billion in another large AI round. Arena's next disclosure of paying customers, or a methodology page for the Alignment Index, would show whether the 31 times is earned.