Arena, the AI leaderboard trusted by millions, hits $100M in annualized revenue

Just eight months after launching its commercial service, AI leaderboard provider Arena — which began as a research project at UC Berkeley in 2023 — has reached $100 million in annualized run-rate revenue. The milestone marks a rapid transition from academic experiment to serious business, driven by demand from AI labs and enterprises for deeper model performance insights.

From crowdsourced rankings to commercial analytics

Arena is best known for its popular crowdsourced AI model performance leaderboard, built from over 10 million user evaluations. Its consumer-facing website lets users type a prompt that is sent to two anonymous models; the user then picks which one performed better. This simple mechanism has made Arena a go-to resource for developers, researchers, and AI enthusiasts tracking the rapidly evolving capabilities of large language models.

While the public leaderboard remains free, Arena began generating revenue in September 2024 with the launch of AI Evaluations — a service that provides model labs and enterprises with deep-dive performance analytics drawn from its community of evaluators. The commercial offering has proven as popular as the free tool, propelling revenue from $30 million in annualized run rate in January to $100 million today.

Revenue model and market positioning

Anastasios Angelopoulos, Arena’s co-founder and CEO, told Bitcoin World that many still view the company as an open-source project. “A lot of people don’t even understand that our business is making any money at all,” he said. While Arena refers to its revenue as ARR — a term traditionally meaning annualized recurring revenue — Angelopoulos clarified that the company charges customers for consumption, meaning revenue is not recurring in the traditional sense.

Arena operates in a unique space. It has no direct competitor in the crowdsourced model evaluation niche; Yupp, another startup in this area, shut down in March. However, Angelopoulos noted that Arena competes “for the same dollar” with human labeling startups like Mercor, Surge, and Scale AI. These companies help model makers refine their AI during post-training, a phase that has become increasingly critical as AI providers push for maximum performance.

Why post-training refinement matters

The appetite for post-training services is surging across the AI industry. When Arena announced its $150 million Series A in January at a $1.7 billion valuation, its annualized revenue was $30 million. For context, Handshake’s gross annualized revenue from AI training has nearly doubled since January, climbing from $550 million to nearly $1 billion, according to The Information. Mercor’s annualized revenue also topped $1 billion earlier this year, up from $500 million last September.

These figures underscore a broader trend: as AI models become more capable, the work of fine-tuning and evaluating them has become a high-stakes, high-growth business. Arena’s platform ranks models across text, coding, vision, and image generation tasks, as well as complex, long-running workflows through its recently introduced Agent Mode.

Founding team and funding

Arena was co-founded by Angelopoulos and Wei-Lin Chiang, a fellow UC Berkeley postdoctoral student who serves as the startup’s CTO. Ion Stoica, the renowned UC Berkeley professor and Databricks co-founder, advised the project before it incorporated as a company in April 2025. Arena has raised a total of $250 million from investors including Felicis, Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners, Laude Ventures, and UC Investments.

Conclusion

Arena’s rapid revenue growth signals that the market for independent, community-driven AI model evaluation is expanding fast. As AI labs race to improve their models, services that offer transparent, crowdsourced performance data are becoming indispensable. Arena’s journey from a Berkeley research project to a $100 million revenue business in under two years reflects the broader acceleration of the AI infrastructure ecosystem.

FAQs

Q1: What is Arena and how does it work?
Arena is a crowdsourced AI model leaderboard that lets users compare two anonymous models side by side. Users vote on which performs better, generating millions of evaluations that rank models by capability across tasks like text, coding, and image generation.

Q2: How does Arena make money?
Arena generates revenue through AI Evaluations, a commercial service launched in September 2024 that provides AI labs and enterprises with detailed performance analytics based on community evaluations. The company charges for consumption, not recurring subscriptions.

Q3: Who are Arena’s competitors?
Arena has no direct competitor in the crowdsourced model evaluation space. However, it competes for the same customer budget as human labeling startups like Mercor, Surge, and Scale AI, which help AI companies refine their models during post-training.