Gemini2.5 Pro Programming Ability Test in Cursor

I have reviewed the tool Agrment code before, and it is very convenient to use. The most important thing is that it can be natively integrated with vs code, which is something that Cursor cannot mention.

I have always heard that Gemini2.5 has great programming capabilities, but it is very expensive to use; so I have never tried it. As a result, I found a Gemini2.5 trial invitation in my gmail mailbox today.

If you don't use this, it will be a disrespect to Google.

According to the official introduction, Gemini 2.5 Pro ranks first in the WebDev Arena ranking of web benchmark test, surpassing Anthropic's Claude 3.7 Sonnet; it supports automatic matching of design styles to maintain consistent style; so programming is its strong point, then I will use programming to test it.

Let's use Cursor to test this model. I used the free Chatgpt version before, but I felt that the programming was very poor. This time I want to test the so-called most powerful programming model.

Preparation

1. 1. Download the latest version of Cursor. Only the Gemini 2.5 Pro version will be executed in the new version of Cursor model list.

1. 1. You need to apply for an APIKEY on Google AI and then fill it into Cursor.

• • Visit  Google AI Studio ^[1] and register/log in
• • Create a new API key (make sure to select Gemini 2.5 Pro access)
• • In Cursor, open Settings and select the “API Keys” tab
• • In the “Google AI (Gemini)” section, enter your API key
• • Return to the "Models" tab and select the "Gemini 2.5 Pro with API Key" option
• • Adjust the desired parameters (temperature, context window, etc.)
• • Click “Save” to apply the settings

Important: This apikey cannot be configured in mainland China because the region does not allow it.

use

I randomly found a screenshot of a web page and gave it to him, and asked him to help me implement a front-end project based on this page; but it was still very inconvenient to use.

• • It can only use the ask mode, it will only answer questions, it will not automatically create files and write code. You need to click run for each step

I found it difficult to ask it to write a project from scratch, so I changed my mindset to learning existing project code and raising issues for it to solve.

I sent it a bug issue of an open source project and asked it to help me find the problem, but after many rounds of conversation, it still didn't mention the key points; overall, it was very average and not impressive.

I used the same bug issue and used  Augment  code to analyze and find the problem. It can clearly analyze the problem and provide a solution.

So I didn’t find its strength in the model, and  Augment Code  uses the Sonnet 3.5 model, which is even more powerful.

Augment  code also has an Agent auto mode, which can automatically control the terminal, create files, write code to files, and perform a series of operations such as dependency tracking and running;

Since I discovered this project last time and have been using it until now, using  the Cursor  function again is really like a manual gear, while  the Augment  code is an automatic gear. The user experience is very different.

I would like to recommend this tool again: https://www.augmentcode.com/. It is really useful when installed on vscode, but if it is company code, you should use it with caution. I use it on open source projects, so there is no problem.

Programming has entered a new world again, from the previous assisted driving Github copilot to the autonomous driving Ai Agent Auto mode, which can be operated fully automatically.

Let us witness the arrival of a great era!!