Alibaba AI Spring Festival "Deification": 130 million people flood into Qianwen, daily active users match Doubao, B2B model price is only 1/18 of Google's

This Spring Festival, the AI industry was “unresting.” Chinese AI companies such as Alibaba, ByteDance, Tencent, and Moonlight Shadow launched an unprecedented Spring Festival offensive.

As of the early morning of February 17 (the first day of the Lunar New Year), the battle was initially decided. Alibaba leveraged its “chip + cloud + model” vertical integration advantage to achieve a dual victory in both B2B and B2C sectors.

In the B2B sector, Alibaba open-sourced a new generation of large model Qwen 3.5-Plus on New Year’s Eve, which not only rivals Google’s Gemini 3 Pro in performance but also offers a market-shocking 1/18 “floor price.”

On the B2C side, the competition was even fiercer. Latest data shows that during the Spring Festival, over 130 million people flocked to the Qwen app to experience AI shopping. Qwen’s daily active users (DAU) soared to around 73 million within just three months, quickly catching up with ByteDance’s “Doubao,” which had accumulated three years of user base.

Behind this Spring Festival battle is a shift in AI value focus—from Chat-based “companionship” to Agent-based “task execution.” Alibaba is attempting to utilize its unique “Tongyun Ge” full-stack capabilities to solve two major industry challenges: Moore’s Law-inspired reasoning costs and the “last mile” of commercial monetization.

B2B “Cost-Reduction Miracle”:

Performance comparable to Google’s flagship, cost only 1/18

On New Year’s Eve, February 16, Qwen 3.5-Plus was officially open-sourced. After the large model “battle of hundreds of models” entered deep waters, Alibaba shifted focus from parameter count to “unit compute efficiency.”

The Qwen 3.5-Plus released by Alibaba adopts an innovative sparse mixture of experts (MoE) architecture, with a total of 397 billion parameters, but only 17 billion active parameters. This “small wins big” architectural innovation directly results in an exponential increase in inference efficiency. Data shows its maximum inference throughput increased by 19 times, and memory usage decreased by 60%.

Even more impressive is inference efficiency. The API price for Qwen 3.5-Plus is as low as 0.8 yuan per million tokens. At comparable performance levels, this price is only 1/18 of Google Gemini 3 Pro.

Alibaba’s goal with this move is to accelerate enterprise AI penetration through extreme cost-performance, while also delivering a “blow” to intermediate model vendors relying on API resale or lacking underlying compute optimization.

Since 2023, Alibaba has open-sourced over 400 models, with global downloads exceeding 1 billion. Its strategy is clear: use open source to spread costs via scale effects, and build an ecosystem moat through low prices.

Birth of a National-Level AI Assistant:

1.3 billion people “awaken” the Agent

Latest data from February 17 shows that during the Spring Festival, users issued 5 billion “Qwen help me” commands to Qwen, and 130 million people experienced AI shopping for the first time. This indicates that “Qwen help me” has become a new user habit in the AI era, and Qwen has quickly become a national-level AI assistant.

The granular growth data reveals astonishing penetration:

Ticketing boom: with the release of Spring Festival movies, Qwen’s movie ticket orders increased by 372 times month-over-month, and transportation ticket orders grew over 7 times;

Downstream markets: AI orders from third- and fourth-tier cities surged by 782 times, with nearly half of all AI orders coming from counties;

Bridging the gap: nearly 4 million users over 60 experienced AI shopping, proving that “one sentence order” is helping to bridge the digital divide.

QuestMobile data shows that on February 7, Qwen’s DAU reached 73.52 million, approaching Doubao’s 78.71 million. Doubao took three years to reach this level, while Qwen achieved it in just three months.

Behind this is Alibaba’s ecosystem “group army” operation. Currently, core assets like Taobao, Alipay, Fliggy, Amap, and Damai are integrated with Qwen, with plans to launch AI ride-hailing, recharge, and other functions.

Compared to OpenAI and Google, Alibaba has a “full-cycle” performance network—from intent to execution—which is currently the most coveted capability among Silicon Valley giants.

“Tongyun Ge” Triad:

Holding the core compute power definition rights

Why can Alibaba support both the B2B “price war” and the B2C “hundreds of millions of concurrent users”? Its confidence comes from its core technology strategy known as “Tongyun Ge.”

“Tongyun Ge” refers to the golden triangle of the Tongyi Laboratory (models), Alibaba Cloud (infrastructure), and Pingtouge (chips). Market sources indicate Alibaba has decided to support Pingtouge’s future independent IPO.

In the global AI arms race, vertical integration is the only way to reduce costs. Google has validated this with “TPU + Cloud + Gemini,” and Alibaba has gone even further. By self-developing chips like “Zhenwu” and deeply coupling them with the Qwen models, Alibaba has created an integrated “AI supercomputer.”

This underlying architecture unification allows Alibaba to maximize compute efficiency. Under FP8 and FP32 precision strategies, training costs are significantly reduced, and training speed is increased by 10 times.

For the market, Alibaba’s narrative is changing.

In the past, there were concerns that Alibaba lacked a phenomenon-level entry like ChatGPT in the AI era. But this Spring Festival’s results show that Alibaba is forging a different path from Silicon Valley—driving forward with “the strongest open-source models + the most comprehensive physical ecosystem”: not just creating chatbots, but building capable Agents.

As AI moves from “trial” to “practicality,” Alibaba, with its compute pricing power and commercial fulfillment capabilities, may just be beginning to reveal its true strength in the AI era.