Just eight months after launching its commercial service, Arena—the AI leaderboard provider that started as a research project at UC Berkeley in 2023—has reached $100 million in annualized run-rate revenue.
Arena is widely recognized for its crowdsourced AI model performance leaderboard, which has aggregated over 10 million user evaluations. On its consumer website, users type a prompt and receive responses from two anonymous models, then vote on which one performed better. This simple mechanic has turned Arena into a trusted, community-driven benchmark for the AI industry.
From Free Tool to Revenue Machine
While the public leaderboard remains free, Arena started generating revenue in September 2024 with the launch of AI Evaluations. This service offers model labs and enterprises deep-dive performance analytics derived from the platform's active evaluator community. The rapid revenue growth signals strong commercial demand—matching the enthusiasm seen from the community, which frequently joins for early access to cutting-edge, often unreleased, AI models.
“A lot of people don’t even understand that our business is making any money at all; people still see us as like an open-source project,” said Anastasios Angelopoulos, Arena’s co-founder and CEO, in an interview with TechCrunch.
Clarifying the Revenue Model
Although Arena describes its milestone as ARR—short for annualized recurring revenue—Angelopoulos clarified that the company charges customers based on consumption, not subscriptions. This makes the revenue technically non-recurring, yet the scale reached within months underscores the platform's value proposition.
A Unique Competitive Landscape
Arena operates in a space with few direct competitors. Yupp, another crowdsourced model-picking startup, shut down in March 2026. Still, Angelopoulos notes that Arena competes for the same budget as human-labeling firms like Mercor, Surge, and Scale AI. These companies help AI developers refine models during post-training—a critical phase where Arena’s crowd-sourced evaluations offer a complementary or alternative approach.
Demand for post-training refinement services continues to surge across the AI industry. Arena’s January 2026 announcement of a $150 million Series A round at a $1.7 billion valuation cited $30 million in annualized revenue at the time. For context, Handshake’s gross annualized revenue from AI training nearly doubled from $550 million in January to nearly $1 billion by April 2026, as reported by The Information. Mercor’s annualized revenue also crossed $1 billion earlier in 2026, up from $500 million in September 2025.
Expanding Benchmarks and Leadership
Arena ranks models across a variety of tasks, including text, coding, vision, and image generation. It also evaluates complex, long-running workflows through its recently introduced Agent Mode, reflecting the industry’s shift toward autonomous AI agents.
The company was co-founded by Angelopoulos (pictured left) and Wei-Lin Chiang (pictured center), both former UC Berkeley postdoctoral researchers. Chiang serves as CTO. The startup’s journey from academic project to a $100M revenue business in under three years illustrates the growing commercial appetite for transparent, community-driven AI evaluation in 2026.
via TechCrunch AI