TENCENT(HK:00700) 华尔街见闻·2026-02-12 09:55Core Viewpoint - The 2026 Spring Festival marks a significant shift in the competition among Chinese AI giants, focusing on the "efficiency" and "intelligent agent" implementation of flagship models rather than just model performance [2][10]. Group 1: Model Releases and Competition - The Spring Festival has become a crowded "release season" with multiple flagship and near-flagship updates from various companies, unlike previous singular releases [3]. - ByteDance leads with a trio of models: Seedance 2.0 (video), Seedream 5.0 (image), and Doubao 2.0, with Seedance 2.0 showing signs of becoming a "hit" [3]. - Alibaba is set to launch Qwen 3.5 in mid-February, supported by a 3 billion yuan incentive plan to attract users [4]. - Zhiyu released GLM-5 on February 11, expanding its parameter scale from 355 billion to 744 billion [5]. - DeepSeek is expected to unveil version V4 in mid-February, focusing on improvements in encoding and handling long prompts, with support for up to 1 million tokens [6]. Group 2: Market Dynamics and Implications - The simultaneous release of multiple models will lead to intensified comparative testing, making it crucial for developers to present credible flagship updates to avoid being dropped from trial lists [8]. - The Spring Festival is viewed as a reset period for user preferences, where users will experiment with various products but quickly decide which to continue using [9]. - DeepSeek's potential release is anticipated to have a significant impact on platform economic benefits rather than just the chatbot itself [10]. Group 3: DeepSeek's Technological Advancements - DeepSeek's latest paper reveals a technological path that enhances quality without heavy computational upgrades, using "conditional memory" as a second sparse axis [11]. - If implemented as described, this could lead to efficiency improvements, allowing AI to be economically embedded in high-frequency consumer products rather than remaining standalone chatbots [12]. Group 4: Beneficiaries of the Model War - Surprisingly, the biggest beneficiaries of the model war may not be the model vendors but Tencent, which owns the high-frequency communication interfaces WeChat and QQ [13][14]. - Tencent is expected to integrate third-party model capabilities into its core consumer interfaces, enhancing user experience [15]. - For Alibaba and Baidu, stronger models could improve user experience but may also face pressure from potential price wars initiated by DeepSeek [16]. Group 5: Market Sentiment and Future Outlook - Despite the enthusiasm in the capital markets, there is a cautious perspective regarding the actual implementation of consumer-level AI, with large-scale user testing during the Spring Festival serving as a critical test [18][19]. - The true signal of adoption will not be the initial release hype but whether existing giants will integrate AI as a default feature in high-frequency interfaces, driving sustained demand for reasoning capabilities [20]. Group 6: Valuation and Long-term Profitability - Zhiyu's GLM-5 has achieved state-of-the-art capabilities in agent functionality, while MiniMax has realized dual commercialization in B2B and B2C through its full-spectrum models [22]. - Morgan Stanley's valuation logic looks beyond short-term losses, projecting towards 2030 profitability, with target prices set at 400 HKD for Zhiyu and 700 HKD for MiniMax based on a 30x expected P/E ratio for 2030 [23][24]. - As model capabilities approach global frontiers, the rationale for valuation adjustments will shift towards economic benefits, including stronger willingness to pay and higher API workload retention rates [25].