New Year's Eve "Double Bang"! Alibaba's Major Announcement

Breaking Announcement

As the lights in thousands of households turn on on New Year’s Eve and families gather for their annual dinner, a silent “nuclear explosion” is happening inside an office in Hangzhou. On the evening of February 16, Alibaba officially open-sourced its long-anticipated next-generation large model—Qwen3.5-Plus—marking a major milestone.

Almost simultaneously, countless young people miles away are shouting at their phone screens: “Qwen, help me”—some are rushing to buy the last movie ticket, others are ordering a hot milk tea for a long-distance lover, and some are planning their return flights.

This is China’s AI battlefield during the 2026 Spring Festival: on one side, deep technical dives into code; on the other, a consumer frenzy of fireworks and festivities.

Model Open-Source on New Year’s Eve

The 2026 Spring Festival sees unprecedented “gunpowder” in China’s AI “battlefield.” On February 16, New Year’s Eve, while most people are immersed in the joyful atmosphere of family reunion, Alibaba quietly dropped a “deep water bomb”—open-sourcing the new generation large model Qwen3.5-Plus.

The released Qwen3.5-Plus has a total of 397 billion parameters, with only 17 billion active. Thanks to its architecture innovation of “small wins big,” it outperforms the trillion-parameter Qwen3-Max model, with 60% less deployment memory and up to 19 times higher maximum inference throughput. The API price for Qwen3.5-Plus is as low as 0.8 yuan per million tokens, just 1/18 of Gemini 3 Pro.

In multiple public evaluations, such as the MMLU-Pro knowledge reasoning test (87.8 points) and the GPQA doctoral-level difficult questions test (88.4 points), Qwen3.5 outperformed closed-source models like GPT-5.2 and Claude 3.5.

This marks the third consecutive year that Qwen has released major updates on New Year’s Eve. From Qwen2.5 to Qwen3, and now to Qwen3.5, this “champion of innovation” team continues to release new technical versions during traditional festivals.

It is understood that most existing multimodal models still rely on stacking modal data and parameters. Alibaba not only employs a hybrid MoE architecture but also uses a unified architecture during pretraining to handle multiple modalities simultaneously. With 397 billion total parameters and only 17 billion active, it is one of the smallest models with comparable performance.

In fact, just before the release of Qwen3.5, Qwen-image-2.0, which combines image generation and editing, quietly debuted and ranked among the top three globally on the AI Arena leaderboard.

130 Million People Using AI for Shopping

If open-source models are Alibaba’s technical foundation, then the application of Qwen APP demonstrates the potential of AI to empower consumer ecosystems. Recent data shows that over 130 million people flocked to the Qwen app for AI shopping during the Spring Festival. The app’s daily active users (DAU) soared to around 73 million within just three months, quickly catching up with Douban’s three-year accumulation.

Notably, AI is rapidly penetrating into lower-tier cities. Nearly half of all AI orders come from county towns. Orders for movie tickets from third- and fourth-tier cities surged by 782 times. Additionally, thanks to the extreme convenience of “one-sentence ordering,” nearly 4 million users over 60 years old experienced AI shopping for the first time.

This “全民AI生活方式演习” (nationwide AI lifestyle drill) quickly gained attention. On February 12, Qwen released data showing that within six days, users issued 4.1 billion “Qwen, help me” commands, and AI completed over 120 million orders.

“Qwen aims to integrate AI into everyday life scenarios of ordinary people. This is an inevitable future,” said Wu Jia, President of Qwen’s consumer business group. “AI is transforming from a chat tool into a capable assistant.”

On February 14, Qwen announced an additional investment of 3 billion yuan to expand AI shopping, extending the scope of its super free-order cards from milk tea and New Year goods to movie tickets, attraction tickets, flights, and hotels. The Apple App Store’s free ranking shows Qwen App has been number one for several consecutive days.

Wu Jia stated that the first wave of Spring Festival activities was a capability validation—testing whether AI could run the full “understanding—decision—ordering—payment” chain in real-life scenarios. The second wave involves Alibaba’s entire ecosystem, including Fliggy’s flights and hotels, Damai’s movie tickets, and soon, Gaode’s group buying and ride-hailing services.

Behind the Scenes: “Tongyun Ge”

Qwen’s rapid iteration and deployment are closely related to its underlying technical architecture’s completeness.

In late January, market rumors indicated Alibaba supported chip company Pingtouge’s independent listing. This signaled the emergence of the “Tongyun Ge” team, composed of Tongyi Laboratory, Alibaba Cloud, and Pingtouge. With this layout, Alibaba has become the only domestic and one of the few global companies capable of independently developing AI chips, cloud platforms, and large models.

This strategic formation has been years in the making. In 2017, Alibaba established DAMO Academy to explore cutting-edge technology; in 2018, Pingtouge Semiconductor was founded and began developing large models. Before the AI boom, these investments were seen as “forward-looking,” but now they underpin Alibaba’s technological competitiveness.

Globally, tech giants have different focuses in AI. OpenAI leads in large model technology but lacks capabilities in payments and logistics; Google and Microsoft have operating systems and office ecosystems but mainly serve online; Amazon has a strong logistics network but is relatively lagging in large models.

In contrast, Alibaba combines influence in open-source models with a nationwide application matrix—Taobao, Alipay, Ele.me, Fliggy, Gaode—covering high-frequency scenarios like e-commerce, local services, travel, and hospitality, forming a complete loop from model capability to service fulfillment. The phased achievements of Qwen are a concentrated release of this ecosystem in the AI era.

It’s worth noting that Alibaba’s exploration of AI shopping scenarios has already prompted overseas follow-up. On February 11, Google announced integration of shopping features into Gemini, allowing users to complete transactions directly within AI chat interfaces. This indicates that Chinese AI companies are leading the way in commercializing agent-scale deployments, pushing global tech giants to accelerate their AI strategies.