AI Front End Generator Comparison: Claude Code vs. v0 vs. Cursor vs. Replit (www.hansreinl.de)

A hands‑on comparison tested several emerging AI front‑end generators (Lovable, Replit, Vercel v0, base44, Cursor, GitHub Copilot via VS Code, and Claude Code) by giving each the same prompt and feature list to build a greenfield web MVP (“Speakit”). The goal was to measure real-world readiness using objective metrics (Lighthouse performance/accessibility/SEO, code quality like LOC and cyclomatic complexity, exportability, tech stack, Git integration, error handling, cost) plus subjective developer experience and iteration speed. The exercise frames a broader industry shift toward “vibe coding” — describing interfaces to an LLM and iterating — which could speed delivery while raising questions about maintainability and accountability. Key findings: outputs varied widely. Lovable, Vercel v0 and Claude Code produced top performance and accessibility (Lovable mobile/desktop Lighthouse 98/100; v0 92/93; v0 and Claude hit 100 accessibility/SEO in places), modern TypeScript/React stacks, and better export/Git support (Lovable and v0 offered Git downloads). Replit’s dev build lagged on performance (~54/55) despite rich features. Cursor, Copilot and Claude produced compact code but often lacked .git export and test coverage; most tools generated no tests and limited error handling. Costs clustered around ~$20–25/month plus usage. The experiment shows these tools can rapidly produce deployable frontends, but code hygiene, long‑term maintainability, and collaboration workflows remain uneven — important caveats as teams adopt AI‑driven front‑end generation.