AI News·8 Nov 2026·via Xda-developers

    I made Claude Code, Codex, and Antigravity build the same Chrome extension, and only one was a keeper

    An individual test of AI coding tools for Chrome extension development found Claude Code, utilizing the Opus 5.5 model, delivered the most effective outcome. The comparison, which included Google Antigravity and Codex, tasked each tool with building a complex, offline screenshot utility, revealing significant differences in feature implementation and user experience.

    A practical comparison tasked Claude Code, Codex, and Google Antigravity with building a specific Chrome extension, SnapNest. This offline screenshot and beautification tool required multiple capture options, an editor, customizable backgrounds, annotations, various export formats, and keyboard shortcuts. A key requirement was strict privacy, demanding minimal permissions and no network requests.

    Claude Code, powered by the Claude Opus 5.5 model, was reported to provide the best experience. Almost all requested features functioned perfectly, and its editor layout was described as well-thought-out with easy-to-find controls. Notably, it addressed usability issues like small numbered annotations by offering better sizing options, and included a useful preset system for saving styles. This comprehensive implementation and refined user interface distinguished its performance.

    In contrast, Codex, using the Luna 6 model, built a functional extension but suffered from an unappealing interface, tiny fonts, and awkward side-panel controls that complicated editing. Google Antigravity, leveraging the Gemini 3.8 Flash model, offered a better layout and useful export functionality for formats like PDF and WebP. However, its annotation process was less user-friendly, relying on JavaScript prompt dialogs, and its blur effect lacked polish.

    Share this article

    Go deeper
    Original reporting by Xda-developersWe don't republish, read the full story →

    Related reading

    6 stories