The battle among tech giants in the AI sector has intensified dramatically. Pony Ma, at Tencent's annual meeting, made a rare public reflection, candidly admitting that "AI progress has been too slow," and commented on "Qwen's full integration into the Alibaba ecosystem," stating that Tencent would consider its large language models and AI products holistically, attempting to carve out a new space with its "Yuanbao" strategy; Baidu Intelligent Cloud raised its "2026 AI-related revenue growth target" to 200%, making an all-out push to become the leader in the AI cloud market; ByteDance is even more aggressive, with its Doubao phone attempting to directly capture the hardware terminal market, and a second-generation Doubao AI phone is currently in intensive preparation. Just as everyone expected the competition to peak after the Spring Festival, late on January 26th, without even holding a press conference, Alibaba released its most powerful model yet. The Qwen flagship reasoning model, Qwen3-Max-Thinking, was officially launched. According to disclosed data, Qwen3-Max-Thinking is currently Alibaba's largest and most capable reasoning model, with performance metrics comparable to GPT-5.2 and Gemini 3 Pro. It set new state-of-the-art (SOTA) records in several of the 19 recognized major model benchmark tests. Qwen PC and web versions were subsequently seamlessly integrated, supporting deep logical reasoning and self-verification. There is another interesting development. On January 26th, Guoxing Space disclosed the successful deployment of the Qwen3 large model onto the "Star Computing" program's 01 group space computing center. This marks the world's first instance of deploying a general-purpose large model from the ground to an in-orbit satellite and executing end-to-end reasoning tasks. Netizens joked that Qwen has truly "gone to heaven." Since the start of the year, in the crucial battle for the future "super portal," Alibaba has launched a comprehensive offensive, from integrating Qwen into its ecosystem, to plans for its chip subsidiary T-Head to pursue a separate listing, and the release of the most powerful version of the Qwen model. Actions at the chip, model, and application layers have followed one after another. As AI enters the latter half of the game, each company's strategy differs. "Everyone knows AI will be important in the future. Regardless of when they started, major tech firms are building their AI ecosystems, but the direction isn't set yet. Alibaba is building an ecosystem, Tencent is focusing on social, Baidu is betting on the cloud, and ByteDance is working on hardware. It's not yet the time for clear differentiation," said a senior observer of the AI industry. Qwen's reasoning capabilities have evolved further. The Qwen3-Max-Thinking model is officially described as "the strongest domestic AI large model to date, closest to international top-tier models." Its goal is very clear: to implement a series of innovations in reasoning technology, ultimately achieving a significant leap in model performance. According to detailed information released by Alibaba, the model has over a trillion total parameters and set new global records across 19 authoritative benchmarks covering factual knowledge, complex reasoning, instruction following, human preference alignment, and Agent capabilities. In a key enhancement to the model's reasoning ability, the new Qwen model adopts a novel Test-time Scaling mechanism, which improves reasoning performance while being more economical. Industry-standard reasoning computation often simply increases parallel reasoning paths, redundantly deducing known conclusions, leading to inefficiency. The new Test-time Scaling mechanism employed by Qwen can extract "experiential" insights from previous reasoning results and perform multiple rounds of self-iteration based on this, enabling more efficient reasoning computation within the same context and yielding more intelligent reasoning outcomes. Simply put, it's akin to preventing the AI from "using one scratchpad for the entire calculation" when solving a problem. Instead, it can, like a human, review previous thought steps, extract "experience," and more intelligently utilize this experience in subsequent steps, thereby achieving more efficient and accurate complex reasoning. Based on this reasoning innovation, Qwen's reasoning performance and efficiency have seen significant improvements. For instance, in the HLE test, often called the "final human test," Qwen scored 58.3, surpassing GPT-5.2-Thinking's 45.5 and Gemini 3 Pro's 45.8, achieving the highest score among all current models. Ahead of the impending Agent era, Qwen3-Max-Thinking has also significantly enhanced its native Agent capability for autonomously invoking tools. Specifically, after initial fine-tuning for tool use, the Tongyi team further trained the model using joint reinforcement learning based on rule rewards and model rewards on a large number of diverse tasks, enabling Qwen3-Max-Thinking to think more intelligently in combination with tools. This adaptive tool invocation capability can be fully experienced on QwenChat, where the model autonomously selects from three core Agent tool functions—search, personalized memory, and code interpreter—to provide answers at a professional level. Simultaneously, model hallucination has been greatly reduced, laying the foundation for solving real-world complex tasks. Why does the industry place such importance on reasoning and Agent capabilities? Because this directly relates to whether AI can evolve from a "chat toy" into a genuine "productivity tool" or even a "personal assistant." An AI capable of deep reasoning and autonomously using tools is necessary for handling complex tasks like planning trips, writing professional reports, and analyzing data. Perhaps the purpose of Alibaba's "flagship reasoning model" is to move beyond the "dialogue-only" phase and begin preparing for the "super portal" of physical AI. T-Head's independence and Qwen's integration into the ecosystem. Concurrently, Alibaba is making two seemingly opposite moves: spinning off its chip unit while seamlessly integrating Qwen into its ecosystem. T-Head's solo flight is a snapshot for observing the strategic depth of Chinese internet giants' AI strategies. Originating from Alibaba's DAMO Academy, T-Head was initially established to provide proprietary computing power for Alibaba Cloud and the internal ecosystem. Its Hanguang and Yitian series AI chips have become part of the computing foundation of Alibaba Cloud. The market interprets Alibaba's push for T-Head's independent operation and potential listing as an attempt to transform the chip unit from a cost center into a revenue center. It's not just Alibaba; Baidu is also promoting the independent listing of Kunlun Chip, indicating an industry trend. As large model parameters enter the trillion era, the cost of training and inference rises exponentially, making computing power the "water, electricity, and coal" of the AI age. For major internet companies, developing in-house chips can not only optimize performance and reduce reliance on international giants like NVIDIA but also enable deep synergy with their own AI frameworks and models at the architectural level, forming a vertically optimized closed loop from underlying hardware to upper-layer applications. Spinning off the chip company is driven by revenue considerations. It can force technological iteration through external pressure and provide long-term funding for chip R&D via capital markets. The goal is to support their own cloud business on one hand, and on the other, to export technology, establish industry-standard infrastructure, and generate external revenue to fund R&D. After building foundations at the computing power and model layers, Alibaba's ultimate goal is becoming clearer: integrating Qwen into its vast commercial ecosystem to capture the AI "super portal." Recently, Alibaba has conducted a series of紧凑的 ecosystem adjustments. In December 2025, the group consolidated resources to establish the Qwen C-end business unit, directly targeting consumer AI product experiences. On January 15, 2026, the Qwen APP launched a major version update, officially announcing an upgrade from a "chat tool" to an "action era," deeply integrating with core Alibaba ecosystem businesses like Taobao, Alipay, Taobao Quick Deals, Amap, and Fliggy. It enabled full-process AI services for ordering food delivery, shopping, booking flights and hotels, launching over 400 AI action functions at once, connecting the closed loop from demand expression to task execution and payment fulfillment. For example, a user can tell Qwen, "I want to go camping outside the city this weekend, with a budget of 500 RMB per person." After understanding this vague intent, Qwen's underlying "Agent" capability can simultaneously activate multiple plugins: querying suitable campsites and real-time traffic conditions via Amap, booking a campsite via Fliggy or the local services section, recommending and comparing prices for tents and sleeping bags via Taobao, and finally integrating discount packages and completing payment via Alipay. The user doesn't need to switch between multiple apps; the entire process from idea to consumption is completed through natural conversation. In the future, Qwen will continue integrating with other core Alibaba business lines like Youku, Damai, Cainiao Stations, AliHealth, 1688, and Freshippo, further enhancing ecosystem synergy and scenario coverage. In other words, using Qwen as an entry point can seamlessly connect scenarios like "e-commerce transactions," "local life," "entertainment content," and "payment finance." Qwen itself could become an "AI butler" covering various aspects of users' lives. Guohai Securities pointed out in a research report that Qwen's ecosystem integration model redefines the value logic of an "entry point," pushing industry competition from comparing single model capabilities to a full-dimensional comparison of "model + ecosystem + scenario." Divergent paths for major players. Concurrently, other internet giants are also ramping up their offensives. Tencent, as Pony Ma indicated, appears cautious about aggressively releasing general-purpose large models. However, its strength lies in its social product matrix and deep industrial internet foundation. Tencent's AI strategy leans more towards "infiltration" and "integration," deeply embedding AI capabilities into national-level apps like WeChat, QQ, and Enterprise WeChat, as well as its strongholds in gaming and fintech. Tencent prefers a non-intrusive approach, achieving ecosystem integration "as gently as the spring rain moistening the earth." Baidu's first major announcement of the year focused on "AI Cloud." The Ernie large model provides the underlying model capability, with commercialization centered on offering AI solutions. Its business growth and bidding data have shown strong performance. Baidu's concept of an "entry point" is more B2B-oriented, aiming to be the preferred partner for enterprise and institutional intelligent transformation. ByteDance is known for its agility and pragmatism. Doubao is widely used internally, including in TikTok's content recommendation and creation tools, and Feishu's office collaboration. The launch of the "Doubao AI Phone" and development of "Doubao AI Earbuds" indicate ByteDance is beginning to touch the hardware layer. Its potential AI super portal could be TikTok, Doubao, or a yet-to-be-fully-formed new hardware product. In contrast, Alibaba, from T-Head's underlying chips to Qwen's model layer, and further to the Qwen APP upper-layer application and Quark AI glasses hardware, is attempting to build a complete closed loop. The advantage of this model is autonomy, control, and significant synergistic potential. However, the战线 is long, each segment requires massive investment, and it places very high demands on the company's resource integration and strategic resolve. Especially after Qwen announced its full integration into the Alibaba ecosystem, there have been external质疑. Many users expressed, "I don't want to be confined to Alibaba's ecosystem; I just want more choices." This poses a challenge to Alibaba's ambition of making Qwen a "super portal," as user willingness and habits are often the most difficult factors to influence.