Arena hits $100M revenue run rate eight months after commercializing AI model evaluations

Arena, the crowdsourced AI model comparison platform that originated at UC Berkeley, has achieved a $100 million annualized revenue run rate just eight months into its paid offerings. The milestone underscores growing enterprise demand for third-party AI performance evaluation services as the industry races to refine model capabilities.

From Research to Revenue#

The platform launched its commercial service in September, introducing AI Evaluations—a premium offering that packages insights from its community assessments into actionable analytics for model developers and companies. Despite the rapid monetization, founder and CEO Anastasios Angelopoulos notes that many still view Arena primarily as an open-source initiative, unaware of its commercial operation.

When the company announced a $150 million Series A round in January, its annualized revenue stood at $30 million. The threefold jump in six months reflects how urgently AI labs seek independent benchmarking tools as competition intensifies.

How It Works#

Arena's core product remains freely accessible: users submit prompts that the platform routes to two AI models, then indicate which response performs better. Over 10 million such evaluations have populated its publicly available leaderboard, which ranks models across text, code, vision, image generation, and multi-step agent workflows. This crowd-sourced approach generates the performance data that underpins its commercial analytics product.

The company clarifies that revenue comes from consumption-based pricing rather than recurring contracts, distinguishing it from traditional SaaS arrangements.

Competitive Landscape#

While Arena has no direct rivals—Yupp, a competitor in crowdsourced model evaluation, shut down in March—it competes for the same enterprise dollars as human labeling firms. Companies like Surge, Mercor, and Scale AI similarly help AI teams improve models through post-training refinement. Mercor and Handshake have also seen explosive growth: Mercor's annualized revenue topped $1 billion earlier this year, up from $500 million in September, while Handshake's AI training revenue nearly doubled in the same period to approach $1 billion.

Team and Backing#

Arena was co-founded by Angelopoulos, UC Berkeley postdoctoral researcher Wei-Lin Chiang (CTO), and advised by Ion Stoica, a renowned Berkeley professor and Databricks co-founder, before the company formally incorporated in April 2025. It has secured $250 million in total funding from investors including Andreessen Horowitz, Felicis, Kleiner Perkins, and Lightspeed Venture Partners.