TRAE's Monthly Active Users Surpass 1.6 Million! Let's Talk About the AI Coding Ecosystem Through Data
Trae released their annual developer report.
This can also be seen as an ecosystem survey for AI Coding. After reviewing it closely, I found some interesting data.
TRAE now has a total user base of 6 million and 1.6 million monthly active users. For a relatively professional programming tool, this is quite impressive, possibly number one in China.
Weekly active users on TRAE are almost always present on weekdays, demonstrating extremely high user engagement.
Their iterations over the past few years have been quite rapid. From the initial Agent 1.0, then SOLO Beta, and finally the official SOLO version, I've tested almost every version, and the progress is truly visibly significant, with substantial improvements in usability.
Previously persistent network latency issues and memory usage problems reported by many users have now been greatly improved.
According to their official report: completion latency has been reduced by 60%, and memory usage by 43%.
Regarding usage scenarios, the data distribution is as follows:
1. Code generation: approximately 30%
2. Bug fixing: approximately 40%
These two categories represent the most frequently used tasks and are also where AI models perform relatively well.
As for less frequently used scenarios, repository management, environment management, and code optimization, these functions are indeed areas where users currently have less trust.
In terms of programming languages, we can look at the ranking of programming languages that AI can currently solve. Vue ranks highest, followed by Python, JS (JavaScript), HTML, Java, and TypeScript, likely still primarily focused on front-end web development.
TRAE also provides excellent support for various AI programming ecosystem tools, currently supporting over 11,000 MCPs.
The most commonly used AI programming mode is the Builder mode (80%), followed by the SOLO Coder mode (40%). Over 80% of users use a mix of AI agents, indicating that users are quite familiar with the strengths and weaknesses of these modes.
With the increasing programming capabilities of domestic models, coupled with TRAE's rich and efficient infrastructure, I feel that there will be significant progress in the field of programming in 2026.
