Study Reveals 2.9x Token Efficiency Gap Between Minimal and Full-Featured Web Frameworks for AI Agents

A new study testing 19 web frameworks reveals a striking 2.9x token efficiency gap between minimal frameworks like Flask and Express versus full-featured ones like Phoenix, with AI agents successfully building functional blog apps in 18 of 19 frameworks — findings that carry major cost implications as agents scale to hundreds of daily tasks.

• A study testing 19 web frameworks finds that minimal API frameworks like ASP.NET, Express, and Flask are significantly more token-efficient for AI agents, clustering between 26-29k tokens, while full-featured frameworks like Phoenix consume up to 74k tokens — a 2.9x gap.
• AI agents successfully build a working blog app in 18 out of 19 frameworks on the first attempt, and complete a follow-up feature addition with an 18/19 success rate, highlighting how capable modern agents are across diverse ecosystems.
• When adding features to existing code, most frameworks converge to a similar 15-30k token cost regardless of initial build expense, suggesting that framework overhead is front-loaded and that a 2.9x efficiency gap becomes increasingly significant as agents perform hundreds of daily tasks.