AI Develops Impressive Minesweeper Clones: Testing Reveals OpenAI’s Codex Ranks Highest, While Google’s Gemini CLI Falls Short

In a recent test by Ars Technica, popular AI coding agents were evaluated on their ability to create a web version of Minesweeper. The agents—OpenAI’s Codex, Anthropic’s Claude Code, Mistral Vibe, and Google’s Gemini CLI—were tasked with including features like sound effects and gameplay twists without human assistance. Codex emerged as the top performer with a score of 9/10, offering robust visuals, chording support, and a unique “Lucky Sweep” gameplay twist. Claude Code followed with a 7/10 for its aesthetics but lacked chording. Mistral Vibe scored 4/10, missing key features like sound and creative gameplay twists. Lastly, Google’s Gemini CLI received a shocking 0/10 as it failed to produce a playable game. This exercise raises questions about resource allocation in AI development, suggesting the results may not justify the ongoing corporate investment in AI technology.

Source link