Hello, I'm Yi'an, specializing in AI programming.
I. Foreword
Yesterday, a friend in the group noticed the "Upgrade to Pro" sign on Trae's overseas version, signaling Trae's official entry into commercialization!
The current plan is $3 for the first month, then $10 per month. This price is half of Cursor's. As for whether its performance is better than Cursor's, you'll see Yi'an's review shortly.
Although the overseas version is paid, the domestic version remains free. Currently, there's no payment entry for the domestic version. Brothers who want to continue using it for free can keep using the domestic version; see the domain in the image below.
Impatiently, I got off work at 11 PM and rushed home to top up $3 to try it out for everyone. The top-up entry is on the official homepage; after clicking "Start Using," you can see the Alipay payment entrance.
Scan the QR code, bind Alipay's password-free payment, and it was topped up in less than 3 seconds.
As a non-first-batch Pro user, I finally activated the long-awaited paid version of Trae!
Official Pro users get 600 fast calls and unlimited slow calls, which is generous compared to Cursor's 500 fast calls. It's truly a benevolent provider that doesn't count by tokens.
This was yesterday's update: besides the Pro version, the core improvements are enhanced Agent capabilities, optimized multi-turn conversation context execution, and a lower barrier to entry for intelligent agents, making them ready-to-use out of the box.
II. How Does Trae Pro Perform in Programming?
I'll demonstrate with a few case studies. For fairness, both Cursor and Trae used the claude-4-sonnet model in the demonstration process. Don't ask why not claude-4-opus, because Trae doesn't have it, haha.
1) Test Project: Xiaohongshu-like Waterfall Image Display Page
Why choose this?
Everyone is familiar with Xiaohongshu
The effect is intuitive; you can tell good from bad at a glance
Involves common front-end technologies, but not too complex
Here's my prompt:
I want to create a Xiaohongshu-like waterfall image display page with the following requirements:Technology stack:- Use React 18 + TypeScript- Use Tailwind CSS for styling- Use Zustand for state management- Use react-lazyload for image lazy loadingPage structure:1. Top search bar: - Left logo (text "Little Blue Book") - Middle search box with search icon, placeholder "Search content you are interested in" - Right publish button2. Waterfall main body: - Two-column layout (mobile), three-column layout (tablet), four-column layout (desktop) - Each card includes: * Image (100% width, auto height) * Title (max two lines, overflow with ...) * Author's avatar (circular, 24px) and nickname * Like count (red heart icon + number) - Card hover effect: slight lift, deeper shadow - Click card has press effect3. Data loading: - Load 20 items initially - Automatically load more when scrolling to the bottom - Display loading animation when loading - Use mock data, including images of different heightsStyle requirements:- Overall fresh and simple style- Main color: #fe2c55 (Xiaohongshu red)- Background color: #f5f5f5- Card background: pure white, 8px rounded corners- Adapt to iPhone 12 and above modelsInteraction effects:- Pull-to-refresh function- Display default image if image fails to load- Like animation: red heart expands then shrinks on click- Page switching uses fade-in effectPlease generate the complete code, including component splitting, style handling, and data mock.1. Cursor's Performance
I copied the prompt to Cursor, used Agent mode, and didn't need to intervene at all.
It quickly wrote the code, but running npm resulted in an error.
Couldn't understand it? No problem, I tossed it to Cursor to fix it itself.
Unfortunately, after several rounds of fixes, it still reported an error.
2. Trae's Performance
Opened a new window and switched to Builder mode. Indeed, being a paying customer truly pays off; it's genuinely fast with no queuing!
After 5 minutes, waiting for the result, it also automatically ran npm and executed successfully.
Wow! It passed on the first try! This completely defeats Cursor; with the same model, Cursor reported errors and couldn't fix them after several attempts.
3. Video Demonstration
2) Test Project: Carousel Component
Since Cursor's previous demonstration ended in failure, for the second case, let's try something simpler: a common requirement to implement a carousel component.
Here's my prompt:
Help me write a carousel component with the following requirements:Usage scenario: Product recommendation carousel on an e-commerce website homepageSpecific requirements:1. Basic functions: - Automatic playback, switches every 3 seconds - Pause playback on mouse hover - Supports left and right arrow switching - Bottom dot indicators - Supports left and right swipe on mobile2. Style design: - Carousel dimensions: 100% width, 400px height (PC) / 200px (mobile) - Images should maintain a 16:9 aspect ratio, crop excess parts - Left and right arrows: semi-transparent white background, opacity increases on hover - Indicators: white dots, current item is accent color #ff6b6b - Transition animation: fade-in/fade-out effect, duration 300ms3. Technical requirements: - Use Vue 3's Composition API - Native implementation, no reliance on third-party carousel libraries - Supports TypeScript - Uses CSS Modules for styling4. Data structure: Input image array format: [{ id: 1, url: 'image URL', title: 'image title', link: 'click jump link' }]5. Additional features: - Image lazy loading - Supports custom transition time input - Exposes switching methods for parent component calls - Triggers change event on switchPlease provide the complete Vue component code, including detailed comments.1. Cursor's Performance
After 25 calls, an exception occurred.
We had it try to fix it multiple times, but the result was the same.
2. Trae's Performance
Successfully completed without any errors.
Not only did it implement automatic carousel, but it also considered event logging. This time, Trae performed better!
3. Video Demonstration
3) Test Project: Web Crawler
The third project is many people's first web crawler project. Let's see whose code is easier to understand.
Here's my prompt:
I want to write a Python web crawler to scrape information from Douban Movie Top 250, with the following requirements:Target website: https://movie.douban.com/top250Information to scrape:1. Movie ranking2. Movie title (Chinese + foreign)3. Director4. Main actors (max 3)5. Release year6. Production country/region7. Movie genre8. Douban rating9. Number of ratings10. One-sentence synopsis11. Movie poster image linkTechnical requirements:1. Use requests + BeautifulSoup for scraping2. Add reasonable request headers to simulate normal browser access3. Set random delay (1-3 seconds) to avoid too frequent requests4. Handle possible anti-scraping measures (e.g., login, CAPTCHA, if needed)5. Use logging to record scraping progress6. Error handling: network errors, parsing errors, etc.Data storage:1. Save scraped data as a CSV file2. Also save as a JSON file for later processing3. Save image links separately in a text file4. Use pandas for simple data cleaningAdditional features:1. Support resuming scraping (if interrupted, can continue next time)2. After scraping, generate simple data analysis: - Distribution of movies by decade - Number of movies by country/region - Rating distribution histogram - Top 10 directors by number of works3. Generate an HTML report displaying scraped results statisticsCode specification:- Organize code using classes- Add detailed Chinese comments- Follow PEP 8 coding style- Provide usage instructions and dependency listPlease provide the complete code implementation and explain how to handle potential anti-scraping issues.1. Cursor's Performance
I copied the prompt to Cursor again. The project was generated, but it got stuck at dependency installation.
2. Trae's Performance
I copied the prompt with one click and gave it to Trae. It prompted me to install a library in the middle, and I just clicked install.
The final execution showed multiple functions implemented.
Inputting 1 quickly started the scraping, ultimately generating a CSV file, a JSON file, HTML, and a charts report.
This chart is fairly standard.
3. Video Demonstration
III. Real-World Experience Summary
After these three rounds of practical testing, the results are quite surprising. Cursor's strategy is indeed aggressive; many official model contexts are cut in half, and it frequently degrades even after paying. Trae's paid version tested very strongly this time. I know my review is somewhat one-sided, so I'll see after a week of experience how long it can maintain this. However, I believe in ByteDance's strength; finally, there's an IDE in China that can contend with the AI programming leaders.
Performance Comparison
User Experience Details
Response Speed:
Trae Pro indeed has no queuing after payment, and the response speed is visibly improved.
Cursor consumes calls quickly on complex projects.
Code Quality:
Both generate good quality code with detailed comments.
Trae is smarter in handling dependencies and environment configurations.
Both can understand complex Chinese requirements.
Stability:
Trae Pro successfully completed all three projects on the first try.
Cursor's self-healing ability when encountering errors needs improvement.
IV. My Recommendation
Best value for money currently: Trae Pro
$3 introductory price for the first month, more than half cheaper than Cursor.
600 fast calls, 100 more than Cursor.
Unlimited slow calls as a fallback, no worries about running out.
Who is it suitable for?
Programming Beginners: Trae has higher fault tolerance and a high success rate on the first try.
Individual Developers: Affordable price, sufficient features.
Experienced Developers: The $3 trial cost is almost negligible.
V. Concluding Remarks
The competition among AI programming tools is truly fierce. Trae Pro offers an experience comparable to Cursor at half the price, which is definitely good news for developers. However, tools are just tools; choosing what suits you best is most important. If you have a limited budget, you can try Trae Pro. If you prioritize stability, Cursor is still a good choice. If you like freebies, Trae's domestic version remains free.
Lastly, Yi'an has created a Trae discussion group. Interested friends can add the contact information below to exchange usage experiences. You are also welcome to share your experiences in the comment section!
I have prepared a 120,000-character AI Programming Handbook for everyone. Add WeChat: 20133213, remark "AI Programming Handbook", and I will invite you to my AI Programming Exchange Group to discuss the latest AI programming trends together.
I am Yi'an, an AI super individual focused on AI technology research. Every day, I bring you reviews of cutting-edge AI tools and practical experience, interpreting complex technical concepts in an easy-to-understand way. 👇Long press to scan the QR code and follow, let's explore the infinite possibilities of AI technology together!
If you find my article helpful, please give me a like 👍 or heart ❤, so more people with good taste like you can see this content. Thanks 🙏
Recommended Reading
Here's Yi'an's Introduction (Version 2, Let's Be Friends, Limited-time Offer)
My Story with Shenzhen Dachong
PS: This review is a completely factual record, with no paid endorsements. Whether a tool is good or not, you'll know once you try it~