Bolt.new vs V0 vs cursor, which is better
Recently I started helping some founders to build their webapp prototype. And it's actually my first time doing this. So my perspective will be first-hand but not comprehensive.
Plus I'm on free tier for all of those since the reality is there are so many tools out there and paying for $20 each will add up quickly. Also I'm building things part time so I don't want to waste money than the bare minimum I need.
Bolt.new: Great for bootstrapping
Among the tools I've tested, bolt.new stands out as the most impressive, offering excellent bootstrapping capabilities and maintaining user-friendliness even for non-developers.
Dev experience
The standout feature of bolt.new is its ability to generate not just code, but a complete project structure with all necessary files. The platform allows easy project downloads for further modifications and offers one-click deployment capabilities. After evaluating several platforms, I chose bolt.new for project bootstrapping due to its effectiveness and adherence to requirements. It particularly excels in providing clear instructions for external dependency setup. For instance, it generated SQL statements for Supabase table creation, which I simply executed to achieve seamless database integration. Remarkably, I managed to build a fully functional webapp within just an hour through prompting.
Downside
- Free tier credits deplete rapidly
- If you made any modifications to the file, the next prompt will not keep that. For example you need to reset your API keys everytime it generates a new version
- I did specifically told bolt.new to keep the UI style for a version but later on it still alters the style a lot
- When I re-open a project from a day earlier, the previous prompts were gone
- Even deployment is a prompt meaning if you run out of tokens then you cannot deploy
V0: a glorified Claude
Unlike bolt.new, v0 will not give you the project structure, instead it tells you what file to create one by one. I find this pattern is very similar to claude.ai itself. However the fact that I needed to create all the files and copy paste all the content over, makes it less attractive than bold.new, at least for bootstrapping. Also you need to click on different files to have different preview, makes it feel not like a single integrated product.
I did choose Vercel (v0's parent platform) to deploy the code I built with bolt.new. But for v0 itself, I'd rather just use Claude + local dev.
Cursor: most versatile
After bootstrapping my project on bolt.new. I downloaded everything and load it up on Cursor. With Cursor we can easily invoke AI by Ctrl+K, and chat with it. The suggestions are again pretty much what Claude would give, however there is this convenient little "Apply" button that you can click and it goes to your code repo.
I like this feature as it is not a simple wrapper of Claude anymore, it saves me the effort to create/update files.
From a developer's perspective, Cursor emerges as the top recommendation due to its natural fit with daily development workflows. While bolt.new excels at bootstrapping, I prefer a local environment for iterative development rather than relying solely on AI-driven prompts. Although Cursor's free tier limit of 50 prompts isn't sufficient for serious development, the ability to integrate OpenAI or Claude API keys transforms it into a more flexible pay-as-you-go solution.
Which one is worth paying for
Despite not subscribing to any paid tiers, let's analyze their value propositions:
bolt.new pricing
10M tokens for $20/month, this is very poorly presented as in is it 10M tokens per day? per month? And how are the tokens counted? For every LLM api they have different prices for input/output tokens, it's not clear here. Also is the code also counted as token in each prompt? It does generate a lot of code so we need to be careful. On a free tier I feel I got like 5-10 prompts per day, which is actually kinda enough for bootstrapping a new project. But if you want to use it more than bootstrap, you should try the Pro plan.
v0 pricing
This is even worse than bolt.new as in I cannot tell how much is 10-20x higher messaging limit? Also since I didn't feel it is more useful than Claude itself I'd rather just pay for Claude to gain unlimited access.
Cursor pricing
Pro plan offers unlimited slow premium requests, yay! This is the one I feel worth paying for. But I'd also like to compare that with bringing my own API key. What I would do is first plug in my openai and claude API keys, see if my usage can go beyond $20/month. Then upgrade to Pro plan if it's cheaper.
Conclusion
If you are a non-coder, start with bolt.new, it's blazing fast to get things up and running with just prompting. And I'd pay for the Pro plan if I don't code. On the other hand, as a coder, I prefer to leverage the free tier of bolt.new to setup the basics for me, then bring that code to Cursor for future customization. For cursor, I'll bring my own API key to get a feeling of how much I will spend first then make decision about upgrading.