Developers chasing quick prototypes now turn to AI for browser extensions. No more wrestling manifest files or DOM quirks alone. But which model ships code that actually loads? A MakeUseOf test pitted Claude, ChatGPT, and Gemini against a single vague prompt: build a Chrome extension to search Instagram story viewers. Scrolling endless lists to spot one name? Frustrating. The winner would infer needs, handle Instagram’s tricks, and produce something loadable in Chrome.
ChatGPT kicked off strong. Using GPT-5.5 Thinking mode, it spat out a Manifest V3 ZIP in under two minutes. Named ‘IG Story Viewer Search,’ the extension floated a search box over the viewer list. But Instagram lazy-loads users. Initial tries failed. Version 1.4 added ‘Auto-index all’—programmatic scrolling—and ‘Start live capture’ for manual scrolls. Auto mode faltered against Instagram’s defenses. Live mode worked, sort of. Duplicates piled up. Display names split wrong. Counts ballooned to 700-plus for small stories. ‘It solved the original problem in the loosest possible sense… But the indexing was messy,’ the tester noted. Functional? Barely. After multiple debug rounds.
Gemini stumbled hard. No ZIP. Just files: manifest.json, content.js, styles.css. Manual assembly required. The search bar aimed for Instagram’s viewer modal. It never showed. Six fix rounds later—bar appears in round four, but no users found. Abandon ship. ‘I’ve been a huge fan of Google’s AI efforts for a while now, but I keep getting new reasons to be disappointed.’ Gemini lectured on DOM traversal instead of building.
Claude. Three messages total. First version flopped. Then it sniffed Instagram’s DOM, pivoted to internal API endpoints—user’s logged in, after all. Reliable data fetch. Version two listed all viewers in one search bar. Not per-story. One tweak: version three added story thumbnails, timestamps, view counts. Click a story, load its viewers, search clean. Done. No more fixes. ‘Claude built the extension the quickest, with the fewest messages required, and it’s the only tool that managed to build something fully functional by the end.’
This wasn’t toy code. Real-world hacks against dynamic UIs. Claude grasped platform APIs others ignored. ChatGPT hacked around with scrolls. Gemini? Theory, no practice.
Recent tests echo the pattern. A Towards AI analysis from last week ran 30 days of tasks. Claude topped coding: ‘Claude produces more accurate code, catches more bugs on review.’ Scores: Claude 80.9% on SWE-bench, real GitHub issues. ChatGPT trailed at 74.9%, Gemini 65%. A How-To Geek piece eight days ago tested password strength code. Claude dominated: clean, commented, edge-case proof. ‘The clear winner by miles.’ Gemini and ChatGPT lagged on logic.
On X, buzz builds around ‘vibe coding’—prompt a rough idea, iterate fast. Keira from MH Ventures shipped a full extension in 47 minutes with Claude: one prompt, three iterations. ‘The barrier to your first working product has never been lower.’ Others rave: Chrome extensions as AI playgrounds. XDA Developers praised them last month: ‘Chrome extensions are the easiest way to start vibe coding.’ Claude Code tools dominate posts, though rate limits gripe some.
But flaws persist. Reddit devs call Claude ‘best for coding and it’s not even close’ after months of React refactors. A Playcode.io benchmark from January favored Claude on debounce functions: type-safe, JSDoc’d. Gemini fast but sloppy. ChatGPT functional, generic.
Industry shifts. Zapier notes Claude Code as developers’ pick over ChatGPT’s Codex. ‘Claude Code is now the most popular agentic coding tool.’ Benchmarks like SWE-Bench crown it for repo-scale fixes. Yet Gemini edges research; ChatGPT versatility.
For extensions? Claude wins. It reasons like a senior dev—API over hacks. Others prototype, crash. Pros grab paid tiers: Claude Pro, ChatGPT Plus. Free? Gemini tries. But load failures kill momentum.
Vibe coding democratizes builds. A non-coder ships in hours. Gatekeepers? Obsolete. Pick Claude for code that runs. Others for chats.
WebProNews is an iEntry Publication