Comparison Last updated: October 22, 2026 · Checked against vendor docs Oct 2026

Which vibe coding tools are just Claude or GPT wrappers?

Most of them share the same few models. Here's what each one adds on top, so you can pick on what matters.

What each tool runs on, and what it adds

Tool	Model behind it (Oct 2026)	Own model?	What it adds on top	Our take
Lovable	Lovable picks the model and you can't switch it. Its changelog model upgrades have all been Anthropic Claude Opus releases.	No	Supabase backend and auth wired for you, hosting and custom domains, GitHub sync, and a built-in AI connector for your app's own AI features.	Thin model layer, thick product layer. You're paying for the backend wiring and one-click publish, not a better model.
Bolt.new	Standard and Max agents; Bolt chooses the models behind them (Claude Agent has been the default since Bolt v2). A separate Forge agent runs open-source models such as GLM 5.3, Kimi K3 and DeepSeek v4 Pro.	No	A full dev environment running in your browser (WebContainers), Bolt database or Supabase, hosting, and domains.	Wrapper with a real runtime. The in-browser environment is the part you can't get from a chat app.
v0 by Vercel	A "composite model": a frontier base model Vercel swaps as better ones ship, plus its own Quick Edit model and a custom AutoFix model that repairs output while it streams.	Partly (Quick Edit and AutoFix)	Retrieval over current React/Next.js docs and UI examples, error fixing mid-stream, and one-click deploy to Vercel.	More than a wrapper. The custom fix-up models and framework knowledge are why its UI code needs fewer correction rounds.
Replit Agent	Auto routing picks a model per task. Core and Pro users can choose GPT-6, Claude Sonnet and Opus models, or Kimi K3 manually.	None listed	Cloud IDE, built-in databases, hosting and deployments, and 50+ languages.	Multi-model wrapper with the hosting included. Choose it for the all-in-one platform, not the model.
Cursor	Cursor's own models (Composer 2.5, plus the Grok models Cursor ships as first-party). Premium and higher plans add OpenAI, Anthropic and Google models at API list price.	Yes	Its own editor with Tab autocomplete, an agent, cloud agents, Bugbot code review, and Cursor Router for picking models automatically.	Not just a wrapper. It trains its own coding model and builds the editor around it.
Devin Desktop (formerly Windsurf)	Cognition's in-house SWE-2 and SWE-1.7 models, plus Claude, GPT and Gemini. The free tier gets SWE-1.6 only.	Yes	The Cascade-style agent, in-house retrieval (swe-grep) and review (swe-check) models, and hand-off to Devin cloud agents.	Not just a wrapper. Windsurf was renamed Devin Desktop on June 2, 2026, and its own SWE models are the selling point.
GitHub Copilot	A model menu spanning Anthropic, OpenAI, Google, xAI and Moonshot, plus Microsoft's MAI-Code-1.1-Flash.	Small (MAI-Code)	Unlimited completions on paid plans, a cloud agent that turns issues into pull requests, PR code review, and support for most IDEs.	Multi-model wrapper with the deepest GitHub integration. The value is the issue-to-PR workflow.
Claude Code	Anthropic's own Claude models only.	Yes (it's the model maker's tool)	A terminal-native agent for long autonomous loops, plus IDE extensions (including one for Cursor).	The opposite of a wrapper. It's what other tools wrap, sold directly by Anthropic.

How to choose when the models are the same

You don't code and need login, a database and a live URL: the backend wiring matters most, so start with Lovable.
You want to see and run the code as it's written: Bolt.new's in-browser runtime is the difference.
You care most about polished React UI: v0's fix-up models and Next.js knowledge earn their keep.
You already write code daily: pay for the editor and the tuned model, so compare Cursor vs GitHub Copilot and Cursor vs Claude Code cost.
Price check: if two tools run the same model, compare how much usage each plan includes per dollar. That's where wrapper markups show up.

Questions people ask

Are vibe coding tools just ChatGPT or Claude wrappers?

Many app builders, such as Lovable, Bolt.new and Replit Agent, run on frontier models from Anthropic, OpenAI and others rather than their own. Cursor and Devin Desktop train their own coding models, and v0 adds custom fix-up models on top of a frontier base model.

Which vibe coding tool uses its own AI model?

As of October 2026, Cursor (Composer) and Devin Desktop, formerly Windsurf (SWE-2 and SWE-1.7), ship in-house coding models. v0 uses its own Quick Edit and AutoFix models around a frontier base model. Claude Code runs Anthropic's own Claude models.

If two tools use the same model, does it matter which I pick?

Yes. The layer on top decides the result: backend and auth wiring, where the code runs, how errors get fixed, how you deploy, and how much usage you get per dollar.