The race to build the best AI coding assistant just got a proper scoreboard. Arena.ai, the platform formerly known as LMArena, has expanded its Code Arena from a frontend-only prototyping tool into a fullstack development evaluation platform, ranking AI models on their ability to build real, deployable web applications.
As of July 27, 2026, the WebDev AI Leaderboard has accumulated 489,150 votes across 104 AI models.
From prototypes to production-ready apps
Code Arena originally launched in 2025 with a narrower focus: evaluating how well AI models could handle frontend code. Think UI components, layout generation, basic interactive elements.
The fullstack upgrade, which went live on July 2, 2026, is a fundamentally different beast. The platform now supports PostgreSQL databases with authentication and Row Level Security, third-party API integrations, persistent sandboxes with hot reloading, and direct deployments to Vercel.
The evaluation methodology relies on community-driven assessment. Users build real-time web apps using competing AI models, then vote on which model performed better.
Who’s winning, and why it matters
Anthropic’s claude-opus-5-max currently sits atop the leaderboard with a score of 1725. Moonshot’s kimi-k3-max trails closely at 1682.
Arena.ai’s own business trajectory reflects this momentum. The company reportedly hit a $100 million annualized run rate within eight months of launching its enterprise offering, supported by over 10 million monthly users and 82 million total votes across all its evaluation platforms.
The AI infrastructure investment thesis
Arena.ai has no cryptocurrency tokens or blockchain-related assets. The company appears to be pursuing a straightforward enterprise software model.
The leaderboard data also provides a real-time window into which AI labs are pulling ahead in practical coding ability, which has downstream effects on developer tooling markets, cloud infrastructure spending, and the competitive positioning of companies building on top of these models.