Sina Technology reported on the afternoon of October 24th that TENCENT released the "2025 Tencent R&D Big Data Report," which systematically discloses its R&D progress for the first time.

The report reveals that AI has fully integrated into TENCENT's R&D system, becoming a "new engine" driving the improvement of R&D efficiency. Over 90% of TENCENT engineers utilize the AI programming assistant CodeBuddy, with 50% of new code generated with AI assistance. By relying on WeDev to streamline the R&D toolchain, TENCENT has facilitated an overall R&D efficiency increase of over 20%.

In 2025, R&D personnel will account for 76% of TENCENT's workforce, meaning that 3 out of every 4 employees are engaged in R&D. TENCENT generates an average of 325 million new lines of code per month, completing 370,000 requirements, and facilitating 25.2 million deliveries to support rapid business iteration.

Supported by its self-developed mixed-domain large model, AI has penetrated various stages including coding, review, and testing. With 50% of the new code generated with AI assistance, for every two lines of new code written by engineers, one line is completed with AI help. Average coding time has been reduced by 40%, leading to an overall enhancement of R&D efficiency exceeding 20%.

In the code review process, AI involvement reaches as high as 94%, meaning that almost every code submission at TENCENT undergoes pre-examination by an "AI quality inspector" before human engineers intervene. AI directly identifies 28% of code defects, which are then accepted by engineers, resulting in a 44% increase in the effective detection of issues during code reviews, thereby strengthening software quality.

Furthermore, the number of teams utilizing AI for efficiency improvements has grown sixfold year-on-year, with the application scenarios for large models continuously expanding and problem-solving capabilities strengthening.

The implementation of AI is supported by a robust R&D efficiency platform. The report states that with the deep integration and intelligence of WeDev in R&D practices, the daily data intercommunication between efficiency tools exceeds 80 million times, and the platform saves an average of 5.3 million manual operations each month, improving automation by 67% compared to the previous year.

With the support of this efficient platform, the average daily completed requirements in 2025 will reach 16,000, an increase of 25% year-on-year, while the average completion time has been shortened by 12 hours. Specifically, the AnyDev cloud R&D platform has significantly reduced the environment preparation time from one day to just one minute.

In terms of code quality, through a combination of static scanning and AI technology, over 5.4 million code bugs and security vulnerabilities have been fixed throughout the year, with the average time to resolve bugs reduced by eight hours. If a bug appears in the morning, it could likely be resolved before the end of the workday.

On the infrastructure side, the completion of code builds reaches 6.3 million times per week, the utilization rate of testing environments has doubled, and compilation acceleration has saved 900,000 hours cumulatively, equivalent to 102 years.

The efficiency achievements gained from AI and platformization have been validated in TENCENT's business scenarios. The report shows that in independent efficiency improvement activities initiated by TENCENT's R&D Efficiency Technology Committee, 81% of R&D teams have enhanced their overall efficiency with the automated integration of WeDev’s R&D processes.

The WeChat backend team has developed a distributed compilation toolchain, reducing compilation time by 50%; the delivery cycle for WeChat Pay requirements has been shortened by 31%, improving release quality by 14%; iOS end compilation time for mobile QQ has been reduced by 40%; the automation rate of artistic production in TENCENT Games has reached 95%, with the automated 2UV generation tool achieving the standard of manual production; 65% of the new code in TENCENT Cloud comes from the AI code assistant CodeBuddy, resulting in a 31.5% reduction in the bug rate per thousand lines of code; the iteration efficiency of TENCENT Ads has doubled, and 90% of version releases have achieved full process automation; the successful deployment rate of the TENCENT Financial Technology environment has increased to 99%.

TENCENT continues to increase its R&D investment in AI and cutting-edge technologies. According to TENCENT's Q2 report, R&D investment reached 20.25 billion yuan, a year-on-year increase of 17%. From 2018 to the present, TENCENT's cumulative R&D investment has reached 379.5 billion yuan.