Foreword#
Artificial Intelligence is a strategic technology leading the new round of scientific and technological revolution and industrial transformation. The 2025 “Government Work Report” of Mainland China proposed to continuously promote the “AI+” action. In this context, the China Internet Network Information Center (CNNIC) released the “Development Report on Generative AI Applications (2025)”, analyzing the current status of generative AI applications and development in Mainland China and globally.
Chapter 1: Key Characteristics of Generative AI Development#
1. Fruitful Achievements in Mainland China’s AI#
As of June 2025, Mainland China’s generative AI user base reached 515 million, with a penetration rate of 36.5%. By the end of August, a total of 538 generative AI services had completed registration. Notably, products like DeepSeek have gained significant international attention. DeepSeek-R1, with its excellent performance and extremely low cost, has become the world’s fastest-growing generative AI application.
2. Significant Improvement in Logical Reasoning#
The scale of model parameters continues to expand (GPT-4 estimated at 1.76 trillion), and architectural designs are being optimized (e.g., Mixture of Experts, MoE), leading to outstanding performance in multi-turn dialogues, code generation, and complex reasoning tasks.
3. Leapfrog Development in Multi-modal Capabilities#
Transitioning from single-text to any combination of text, audio, image, and video input/output. Models such as OpenAI’s Sora, Kuaishou’s “Keling 2.0”, and Alibaba Cloud’s “Tongyi Wanxiang” have significantly lowered the barriers to content creation.
4. Significant Reduction in Inference Costs and Lightweighting#
Inference costs have dropped by more than 280 times in two years. Companies are launching lightweight models (e.g., “Ziyue 3 Mathematics Model”), pushing AI from the cloud to the edge (mobile phones, IoT devices).
Chapter 2: User Adoption and Demographics#
- Market Landscape: Doubao (72.2%) and DeepSeek (62.0%) lead in user utilization rates.
- Purpose of Use: Answering questions (80.9%), processing text (36.0%), generating images/videos (33.0%).
- Age Structure: Young and middle-aged users are the primary group, with users aged 19 and below accounting for the highest proportion (33.8%).
- Minors’ Application: 15.2% of minor internet users use it to learn extracurricular knowledge.
Chapter 3: Typical Application Scenarios#
| Sector | Application Highlights |
|---|---|
| Agriculture | Shennon large model, intelligent irrigation systems, autonomous farm machinery, digital human live streaming. |
| Manufacturing | “Taipu” digital supply chain, humanoid robot Walker S1, AI quality inspection alliance. |
| Lifestyle Services | AI Search (Search as a Service), intelligent game NPCs, automated code generation. |
| Scientific Research | Lunar science multi-modal large model, Earth science large model “Kunyuan”, solar large model “Jinwu”. |
Chapter 4: Future Outlook#
- Model Integration: Unified system designs like GPT-5 will automatically switch reasoning modes based on needs.
- Open Source Communities: Platforms like GitHub and Hugging Face provide new impetus for technological progress.
- Embodied AI: The combination of robotics and generative AI will lead to more human-like interactive experiences.
- AI Agents: Shifting from “interaction” to “execution,” capable of independently operating computers to complete complex tasks.
[!NOTE] The data in this report is based on a CNNIC survey from June 2025. Read the original report (PDF)

![[Report] Mainland China's Development Report on Generative AI Applications (2025)](/img/news/cnnic_ai_report_2025_cover_hu_3e5d7a045aafcce2.png)