Following the "DeepSeek Moment," China's AI industry enters the "Qianwen Cycle" once again

China’s AI Enters the “Qianwen Cycle”

The Chinese tech industry’s Spring Festival season arrives as expected. Last year during the Spring Festival, the domestic large model DeepSeek made a splash, shocking the international tech community with its innovative results that are good, cost-effective, and open-source, and promoting a breakthrough in AI awareness domestically, known as the “DeepSeek Moment.” This year, China’s AI has entered the “Qianwen Cycle.”

On the eve of the Spring Festival, Alibaba’s “Qianwen” AI family launched a series of moves:

Model Layer: Released the image large model Qianwen-Image-2.0, standing alongside Google’s Nano Banana Pro and GPT Image1.5 in the top tier. Open-sourced the new generation large model Qianwen 3.5-Plus, with performance comparable to Gemini 3 Pro, topping the world’s most powerful open-source models.

Application Layer: Qianwen App launched the “Spring Festival 3 Billion Free Orders” campaign, creating a surge of over 10 million orders in 9 hours. After the free order cards were snapped up, they were “extended by 3 days,” kicking off global AI shopping.

The Spring Festival is a gift from traditional Chinese culture to the tech industry. During the annual holiday, families reunite, three or four generations gather, friends visit each other for tea, creating golden opportunities for new ideas and products to cross generations and social circles. The tech industry has repeatedly proven successful, from Alipay’s “Gather Five Blessings” ten years ago to DeepSeek last year, with a new product often capturing the Spring Festival spotlight every few years.

Where is Chinese AI headed?

This year’s tech Spring Festival season differs from previous years; it is no longer a single “moment” but the beginning of a new “cycle.” After a year of rapid advancement in Chinese AI during the Year of the Snake, industry leaders like Alibaba are stepping beyond parameter counts and traffic metrics, focusing on two fundamental questions: one, what can AI do? two, what should AI do? Combining these questions leads to one core inquiry: where is Chinese AI headed?

On the B-end model side, Alibaba is the leader in China’s AI competition, but it adopts a pragmatic attitude toward model parameters. As early as 2021, Alibaba’s multimodal large model M6 reached 100 trillion parameters, far surpassing Google’s and Microsoft’s previous trillion-scale models, becoming the world’s largest AI pre-training model. However, Alibaba does not pursue “bigness” blindly; it champions open-source initiatives, promoting full-modal and full-size open-source models. To date, Qianwen has open-sourced over 400 large models.

Latest data from the world’s largest open-source community Hugging Face shows that the top three open-source models by adoption rate globally are: Alibaba Qianwen-53%, Meta’s Llama-15%, and Google’s Gemma-14%. Why does Alibaba Qianwen lead so far ahead? Because from the start, it has pursued extreme model capability while prioritizing meeting the practical needs of various social sectors and developers.

Larger models tend to be more capable, but they also have higher deployment barriers, greater computational costs, and longer inference delays. Ultra-large parameter models may not suit all enterprises. For vertical fields like smart hardware and intelligent customer service, smaller models are often more practical. Qianwen’s open-source models range from hundreds of millions to tens of billions of parameters, allowing developers across industries to choose according to their actual needs. This is the fundamental reason for Qianwen’s global popularity. Data shows that Qianwen’s global downloads have exceeded 1 billion, with monthly downloads surpassing the combined total of ranks 2 to 8. Over 200,000 derivative models have been developed based on Qianwen, making it the world’s most recognized open-source model.

Take the newly open-sourced Qianwen 3.5-Plus as an example: it has undergone a comprehensive overhaul of its underlying architecture, with a total of 397 billion parameters and only 17 billion active. It outperforms the 1 trillion-parameter Qianwen 3-Max model, adding significant multilingual, STEM, reasoning, and other data. Its inference, programming, and agent intelligence benchmarks are leading among open-source models, with several metrics rivaling or surpassing top-tier closed models like GPT-5.2 and Gemini-3-pro from the US.

Notably, Alibaba leverages the “Tongyun Ge” full-stack AI capabilities to promote collaborative innovation in AI computing power and models. As a result, the API price for Qianwen 3.5-Plus is as low as 0.8 yuan per million tokens, only 1/18 of the same performance US Gemini-3-pro. Just as low electricity prices make appliances affordable, this extreme cost-performance ratio enables enterprises to afford AI, indirectly allowing consumers to enjoy AI benefits and greatly accelerating AI application adoption.

From “Talking” to “Doing”

On the consumer side, Alibaba emphasizes the “capable of getting things done” concept, meaning it does not prioritize traffic as the primary goal but holds higher expectations for AI’s value creation. The predecessor of Qianwen App was Tongyi App, launched in 2023, but it was not heavily promoted at the time. Over the past three years, many companies have invested heavily in consumer AI products to boost installation numbers, but Alibaba remained patient, holding back and not rushing to follow. It wasn’t until January this year that the revamped Qianwen App, with over 400 capabilities, was unveiled, aiming to push Chinese AI assistants from “talking” to “doing.” Among these capabilities, AI shopping is the most perceptible and popular.

Alibaba believes that “talking” is certainly valuable for AI, but “doing” is essential. “Talking” AI provides entertainment value, but the greater value of AI lies in driving productivity transformation and creating experience upgrades across work, study, consumption, and public services. For example, in AI shopping, users only need to say “one sentence,” and AI can intelligently access commercial resources to precisely match their needs. AI bridges the fastest route between demand and supply, capable of identifying fuzzy, unarticulated needs and uncovering hidden demands, offering vast potential for innovation and consumption growth.

What else can AI shopping do?

What we see today in AI shopping is just its early prototype, much like the first-generation Taobao over 20 years ago, which inevitably had many imperfections. AI shopping is also just one of the over 400 capabilities of the Qianwen App. But what matters is the starting point and the original intention behind it. AI is regarded worldwide as the crown jewel of technology. Besides image and video generation, what else can this cross-era technology do? What should it do? These are questions the industry must answer. Now, Qianwen is focusing on answering these two questions, and we can call this long-term process the “Qianwen Cycle.”

For Chinese tech companies, the pursuit of technology must reach for the stars, while application must serve the real world. This is the “heaven” and “earth” of AI—both are indispensable and rooted in the land. Sunlight shines from above, dew falls from the sky, but ultimately they nourish the land, warming and benefiting humanity. The “Qianwen Cycle” is this process of sunlight warming the earth and dew benefiting all things, with the ultimate goal of nurturing all things to grow strong. Enabling small enterprises to use advanced yet affordable models, and allowing the elderly who can’t type to order breakfast with “one sentence,” are just the starting points of the “Qianwen Cycle.” The ultimate vision of an intelligent society may far exceed our imagination.