XNews

Codex vs Claude Code: 8-Website Front-End Build Test Shows Codex Wins 5-1

Codex vs Claude Code: 8-Website Front-End Build Test Shows Codex Wins 5-1

A controlled experiment pitted Claude Code against OpenAI's Codex on eight identical website builds with the same prompts, brand assets, and copy. Codex won five tasks, Claude won one, and two were ties. Across all tasks, Claude Code used 25 sub-agents, took ~14 hours, output 3 million tokens, and cost ~$444; Codex used 9 sub-agents, took ~5 hours, output 550k tokens, and cost ~$100—a 5.5x token and 4.4x cost gap. Claude's only win came on a camping gear site with smooth single-page scrolling. On an open-ended creativity task, Codex finished in 8 minutes for $1.50 while Claude took nearly 3 hours and $50. Ultra-specific prompts produced ties but Codex remained faster and cheaper. Claude consistently suffered from text overload and layout bugs.

Source posts