Recommended Tools for Vibe Coding
- Essential Gemini
Actually, back when Gemini 2.5 was released, it became my most frequently used AI. The main reason is that it speaks in plain language, meaning when you ask it something, it responds in the most straightforward and understandable way. At that time, I also liked comparing several models, while GPT and Claude often used more complex terminology. When you ask a question, the answer might contain several technical terms you don't understand, leading to further questions.
For someone who doesn't understand coding and wants to use vibe coding, Gemini 3 is the best choice. Personally, I don't recommend asking AI questions directly in coding environments, as these IDEs have template prompts behind them, making it inappropriate. Additionally, too many fragmented questions can occupy the context window, which can lead to issues while coding.
Of course, Gemini 3 is very powerful; besides coding, it can also be used for daily conversations and learning knowledge, making it a great AI for everyday dialogue. Moreover, Google is quite generous, offering a year of pro membership with a simple student verification.
- Vibe Coding Tools
This year has seen a lot of excitement in vibe coding, with great products emerging one after another. Below are my recommendations tailored for different situations:
1) Cursor
If you're just starting to try vibe coding, Cursor might be the most suitable for beginners. It is the pioneer of vibe coding, and its functionalities are quite good across the board. The prompts and tool adjustments are also well done. Whether you're working on simple or complex projects, the quality is still quite decent. Recently, it seems to have accelerated its update pace due to competitive pressure.
Cursor's payment system is also very user-friendly, allowing membership purchases through Alipay in China. After binding your payment, you get 7 days of pro usage for free.
Cursor is my main tool because I'm not someone who works on many projects. Instead, I continuously optimize a single project. Additionally, many of the products I work on require data analysis, needing adjustments based on actual data conditions. Therefore, Cursor is the most suitable.
2) Antigravity
Antigravity is a new product from Google, and its advantage is that it allows free use of the Gemini 3 and Claude 4.5 models. If you previously activated pro, you can also use it here. The front-end interface of Gemini 3 is outstanding, and I use Antigravity to modify project front-ends.
Although it was developed by the previous Windsurf team, there may still be some potential issues with this new product. However, I personally hope it performs well, as Google's pricing should be quite attractive.
3) VS Code + API
This solution is rarely recommended by KOLs. Its biggest advantage is that it's free. For students or those who are temporarily short on cash, it's worth considering.
VS Code is an open-source software and is the parent of most IDEs. Cursor and Antigravity are developed on its foundation. You can install AI plugins that support custom API in VS Code, such as Cline, etc.
Then, you can find free AI APIs to call; OpenRouter has many free models available. Although their performance may be slightly inferior to the top-tier ones, they are free.
4) Claude Code
Claude Code is produced by Anthropic and was once very popular. Although its advantages are not as pronounced now, it remains a top-tier option.
Claude Code essentially uses engineering methods to improve the overall quality of code. It is more suitable for complex, complete projects (though it has many other uses). It natively supports Claude series models, which are relatively expensive and not very friendly to Chinese users. However, you can easily switch to other models using third-party tools like CC Switch. Now, many large model companies in China have adapted Claude Code, offering ample capacity at competitive prices, and it's very convenient to set up.
5) Codex
Codex is developed by OpenAI and is comparable to Claude Code. So let's compare them:
1) Both Claude Code and Codex are suitable for complex projects, and their quality is top-notch.
2) In terms of code quality, Codex is slightly better than Claude Code. However, Codex has a drawback: it behaves like a plain, unassuming engineer, quietly completing tasks without much explanation. Most people find it hard to understand the code. Therefore, it suits two scenarios: first, if you are already a coding expert and the AI is just doing the heavy lifting for you; second, if you decide to hand over all modifications to the AI and not participate at all.
3) Although Claude Code is slightly inferior to Codex, it acts like a warm-hearted person, providing clear and thoughtful comments. This makes code review very convenient.
Both Claude Code and Codex support various membership accounts for use, as well as API access. You should avoid the official API due to its high cost. If you use it frequently, consider getting a membership. Although it can be quite a hassle, it's worth it. If you're only occasionally working on a project, you might consider domestic intermediaries. I've used two so far, both employing reverse engineering methods. The downside is instability, but the prices are much lower.
There is now a method that balances all aspects: simply use GPT/Claude, etc., to create project development documentation, then use domestic packages like GLM 4.6 with Claude Code to write code, and use Codex to find vulnerabilities. This ensures quality at a reasonable price.
Alright, I've provided an overview of the overall usage scenarios for vibe coding. If you're interested in a specific product, feel free to leave a message. I can write some tutorials on them.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
