Prompt-to-App Benchmark

We gave four no-code / low-code agents the same one-paragraph prompt:

Build a task tracker app. Users sign up, sign in, add tasks, mark them complete, and delete them. Store tasks in a real database. Deploy it.

No follow-up prompts. The agents had to infer the stack, design the UI, wire auth, and produce a live URL.

Criterion	Weight	Max points
Working deployed URL	30%	30
Auth works	25%	25
Database persists tasks	20%	20
UI quality	15%	15
Cost / speed	10%	10

• Same prompt, no screenshots, no clarification.
• Time measured from prompt submission to a deployable / deployed app.
• We manually tested signup, login, add task, complete task, delete task, and refresh persistence.
• Token counts are not applicable because these platforms abstract the LLM layer; costs are platform estimates.

• Lovable delivered the cleanest end-to-end experience. It chose Supabase, built a reasonable schema, and deployed without manual steps.
• Bolt.new produced excellent code but overshot on UI polish. It kept refining visual details after the core app already worked.
• Replit Agent was the most deployment-native. The app was live in Replit immediately, though the interface was plain.
• v0 generated the best-looking UI components but is not a full-stack one-shot tool. It left backend wiring to the user.

• These platforms change fast. Results may shift within weeks.
• Cost estimates include platform subscription or usage tiers, not just LLM tokens.
• "Pass" means the core flow worked, not that the app was production-ready.

• Lovable: fastest path from idea to live full-stack app.
• Bolt.new: when you want generated code you can later export and own.
• Replit Agent: when deployment and sharing are the priority.
• v0: when you already have a backend and need a polished frontend quickly.