图源：网络 需要说明的是，"蒸馏"是全球部分AI公司训练自家大模型的常用手段，而Anthropic自己，恰恰是"蒸馏"技术的使用者，甚至存在更激进的数据抓取行为。 值得一提的是，Anthropic如今却成了人工智能领域对中国最不友好的公司之一。2025年9月，Anthropic在官方文件中明确封禁对中资企业的服务。其实指 责中国企业进行数据蒸馏，已经成了美国企业惯用套路。就在2026年2月12日，OpenAI向美国国会提交内部备忘录，明确指控DeepSeek通过复杂的混淆手 段，绕过其安全防护对GPT系列模型实施蒸馏行为。 Anthropic这次直接对中国企业"高调指控"，一个重大背景是进入2026年2月，美国AI概念股出现多轮明显下跌，资本市场对人工智能的未来产生了"颠覆传 统商业模式"的担忧。CNN评论也指出，不少所谓大模型，其实只是精心包装过的搜索引擎。 马斯克嘲讽、资本变脸，中美AI大战升级，中国凭什么破局？ 开工第一天，AI圈就上演"大戏"！美国AI独角兽Anthropic突然发难，指控DeepSeek、月之暗面（Kimi）等3家中国AI企业，靠"蒸馏攻击"抄袭技术。 "蒸馏攻击"是什么？先按下不表 ...

智通财经APP获悉，近日，国际调研机构弗若斯特沙利文(Frost & Sullivan，简称"沙利文")联合头豹研究院发布《中国GenAI市场洞察：企业级大模型调用 全景研究，2025H2》。报告显示，2025年下半年，全球AI厂商发布策略显著分化：中国厂商依托规模化投入在开源生态中占据主导地位，海外厂商则持 续聚焦闭源主线，发布频次与覆盖广度相对有限，呈现出"少而影响力集中的闭源发布特征"。2025年下半年，中国企业级大模型日均调用量提升至37.00 万亿tokens，较2025H1的10.19万亿tokens增长263%，实现阶段性跃迁。头部大模型占比均有提升，其中，阿里云千问增幅最多，占比跃升至32.1%，相较 上半年的17.7%几乎翻倍，领先优势扩大，成为最受中国企业青睐的大模型。 闭源明显收敛、开源模型全面爆发，中国成为全球开源大模型创新引擎 2025H2，全球AI厂商发布策略显著分化：中国主导开源节奏，海外强化闭源影响力。中国厂商依托规模化投入在开源生态中占据主导地位，海外厂商则 持续聚焦闭源主线，发布频次与覆盖广度相对有限，呈现出"少而影响力集中的闭源发布特征"。其中，中国闭源基础模型赛道快速收 ...

Core Insights - Alibaba has released its new model Qwen 3.5, which is noted for its significant advancements in architecture and performance, establishing itself as a leader in the open-source model space [1][2] - The Qwen 3.5-Plus model has 397 billion parameters, a decrease from the previous flagship model Qwen 3-Max, which had one trillion parameters, yet it achieves performance levels comparable to Gemini 3 Pro with less than 40% of the parameters [2] - The model utilizes only 5% of its computational resources for each response, resulting in a token cost that is 1/18th of that of Gemini 3 Pro, showcasing its efficiency [2] Model Advancements - Qwen 3.5-Plus incorporates a gating technology that has been recognized at a global AI conference, allowing other tech companies to benefit from its advancements [3] - The model has evolved through various architectural changes, including the introduction of a mixed attention mode that enhances its ability to process information selectively [2][3] - Alibaba's Qwen series is now positioned to compete directly with Google across all modalities, having integrated text and visual data learning from the outset [3] Industry Position - Chinese companies, including Alibaba, are leading the open-source model race, effectively surrounding closed-source models and setting the stage for significant advancements in the industry [3] - The rapid development of models like Qwen 3.5 suggests that the gap between domestic models and the strongest state-of-the-art (SOTA) models is narrowing, with potential for surpassing competitors in the near future [3]

Core Insights - The release of GLM-5 marks a significant advancement in open-source AI, bringing it into the era of long-task capabilities [1] - GLM-5 has demonstrated its ability to perform complex engineering tasks, such as creating a Game Boy Advance emulator from scratch [2][7] - The model has achieved impressive results in various benchmarks, positioning it alongside proprietary models like Claude Opus 4.5 [10][12][18] - The emergence of GLM-5 signifies a shift in the SaaS industry, as it allows developers to create sophisticated applications without relying on traditional software solutions [29] Group 1 - GLM-5 can run code continuously for over 24 hours, performing 700 tool calls and 800 context switches, showcasing its stability and reliability [2][7] - The model's programming capabilities have been validated against established benchmarks, achieving the top score among open-source models [18][20] - Users have already begun to leverage GLM-5 for various applications, including a 3D version of Monopoly and an academic version of TikTok, with multiple apps submitted for App Store approval [24][29] Group 2 - The open-source nature of GLM-5 disrupts the market previously dominated by closed-source models, empowering developers with new tools [20][29] - The performance of GLM-5 has led to concerns in the SaaS sector, with significant stock declines for companies like FactSet and S&P Global as investors reassess the future of software sales [29] - The model's capabilities represent a transformation from AI as a mere assistant to an independent engineer, potentially reshaping the landscape of software development [29]

Core Insights - The release of GLM-5 marks a significant advancement in open-source AI, bringing it into the era of long-task capabilities [2][25] - GLM-5 demonstrates exceptional programming abilities, successfully creating a Game Boy Advance emulator from scratch, showcasing its stability and reliability in complex tasks [3][9][12] - The model has achieved competitive performance, ranking alongside Claude Opus 4.5 in various assessments, indicating its strong programming capabilities and operational stability [15][17] Group 1: Performance and Capabilities - GLM-5 executed over 700 tool calls and 800 context switches while maintaining consistent syntax and accuracy [12] - It has been recognized for its ability to generate complex applications, such as a 3D Monopoly game and an interactive version of Minecraft, demonstrating its versatility [26][35] - The model's performance in the Vending Bench 2 test has positioned it as the leading open-source model in terms of operational capabilities [23] Group 2: Industry Impact - The emergence of GLM-5 signifies a transformative shift in the SaaS industry, as it allows developers to create sophisticated applications without relying on traditional software subscriptions [38][40] - The release has caused market reactions, with significant declines in SaaS-related stocks, reflecting investor concerns about the implications of AI on software sales [39] - GLM-5's capabilities challenge the previous dominance of closed-source models, empowering developers with tools that were once exclusive to major corporations [40] Group 3: Community and Developer Engagement - The open-source nature of GLM-5 has generated significant interest and demand among developers, with many eager to utilize its capabilities [41] - The model's development has become a focal point for the community, with its headquarters attracting attention as a notable location [42] - The ongoing advancements in AI programming, initiated with earlier versions, have positioned GLM-5 as a leading choice for coding tasks in both domestic and international markets [41]

Core Insights - The article discusses the recent launch of the GLM-5 model by Zhipu, which has received positive market feedback, with a stock price increase of 28.68% on its first trading day [4] - The GLM-5 model features significant updates, including an increase in pre-training data from 23 trillion to 28.5 trillion and the introduction of a new "Slime" framework to support larger model scales and complex reinforcement learning tasks [4][5] - The article highlights the evolving consensus in the industry regarding large models, indicating a shift from basic coding to more complex engineering tasks [5] Company Developments - Zhipu's GLM-5 model has achieved state-of-the-art (SOTA) performance in coding and agent capabilities, closely matching the performance of Claude Opus 4.5 in real programming scenarios [5] - The model's agent capabilities enable various applications, including end-to-end application development and general agent assistance, showcasing its versatility [5] - Other models released around the same time include Step 3.5 Flash by Jieyue Xingchen, Qwen3-Coder-Next by Alibaba, and MiniMax-M2.5, indicating a competitive landscape in AI model development [6] Industry Trends - The updates from multiple model manufacturers reflect a focus on inference efficiency, long context, multimodality, and cost reduction [6] - Technologies such as the MoE architecture and FP8 precision are being implemented, significantly lowering the cost of model calls from "yuan" to "fen" and "li" [6] - DeepSeek's recent updates have increased context length support to a maximum of 1 million tokens, a significant improvement from the previous version's 128,000 tokens [6]

Group 1 - Gao Xin Retail is currently unable to contact its Executive Director and CEO Li Weiping, but the board believes this matter is not related to the company's business and operations, and there is no significant adverse impact on the group [1] - The situation is seen as a short-term emotional impact, while the long-term effects will depend on the company's ability to stabilize management and continue its transformation strategy [1] Group 2 - The AI model Kimi K2.5 from OpenClaw has become the most popular model, surpassing others like Gemini 3 Flash and Claude Sonnet 4.5, indicating a significant rise in the competitiveness of Chinese open-source models in the global AI sector [2] - The increase in market share and download rates of Chinese open-source models reflects a shift from "technological catch-up" to "ecological competition" in the AI industry [2] Group 3 - Alibaba has launched the new open-source programming model Qwen3-Coder-Next, which demonstrates performance comparable to other models while significantly reducing inference costs to only 5% to 10% of similar performance models [3] - This development enhances the open-source ecosystem in the AI programming field in China and showcases technological breakthroughs in lightweight and cost-effective AI models [3]

Core Viewpoint - The article discusses the evolution of the capital market's pricing logic for Chinese tech assets, particularly focusing on Alibaba's advancements in AI technology and its implications for market perception and valuation [6][7][13]. Group 1: AI Model Advancements - The release of Qwen3-Max-Thinking marks a significant breakthrough, outperforming global models like GPT-5.2 and Gemini 3 Pro in various evaluations, indicating a leap in performance for domestic AI models [8][10]. - The model's innovative "Test-time Scaling" mechanism allows for more efficient reasoning and self-iteration, enhancing its ability to produce intelligent results [9][19]. - Qwen3-Max-Thinking's capabilities include native agent abilities, enabling it to autonomously utilize tools and adjust its actions based on feedback, which enhances reliability for enterprise applications [20]. Group 2: Market Dynamics and Revaluation - The initial market reaction to Chinese AI advancements was characterized as "emotional repair," with investors hesitant to fully embrace the potential for leadership in AI [7][13]. - As technological gaps close, the revaluation of Alibaba's AI capabilities is becoming a matter of "when" rather than "if" [13]. - The shift from a focus on computational power to intelligent reasoning represents a new growth curve for the AI industry, necessitated by the limitations of previous scaling methods [15][17]. Group 3: Open Source and Global Positioning - The Qwen series has established dominance in the open-source AI ecosystem, surpassing Meta's Llama series with over 200,000 derivative models and 1 billion downloads [22][23]. - China's share of global open-source AI model adoption has risen to 17.1%, overtaking the U.S. for the first time, reflecting a significant shift in the geopolitical landscape of AI [25]. - Notably, even Silicon Valley companies are adopting techniques from Qwen, indicating its competitive edge in specific capabilities [26][27]. Group 4: Comprehensive AI Strategy - Alibaba is uniquely positioned as one of the few companies globally with a complete stack of AI capabilities, integrating computing power, model development, and application [31]. - The company has made significant investments in hardware and infrastructure, with plans to allocate over 380 billion yuan for cloud and AI hardware over the next three years [34]. - The Qwen APP has demonstrated commercial potential by evolving from a simple chatbot to a comprehensive AI capable of executing complex tasks, thus expanding the monetization opportunities in AI [34].

Core Insights - DeepSeek's release of the DeepSeek-OCR2 model has led to a short-term surge in related stocks, with companies like YunSai ZhiLian and Hongjing Technology hitting their upper trading limits [1] - The DeepSeek-OCR2 model utilizes the innovative DeepEncoder V2 method, allowing AI to dynamically rearrange image components based on their meanings, closely mimicking human visual encoding logic [1][6] Technology Advancements - The DeepSeek-OCR2 model breaks the limitations of traditional OCR by improving semantic understanding of images, significantly enhancing recognition accuracy in complex layouts, distortions, and occlusions [6] - In the OmniDocBench v1.5 benchmark test, the model achieved a score of 91.09%, a 3.73% improvement over its predecessor [6] - The model maintains high precision while controlling computational costs, with visual token counts limited to between 256 and 1120, aligning with Google's Gemini-3 Pro [6][7] Architectural Significance - The release of DeepSeek-OCR2 represents not just an upgrade in OCR performance but also a significant exploration of architecture, validating the potential of using language model architectures as visual encoders [7] - The model's "two cascaded 1D causal reasoning" approach may signify a breakthrough in achieving true 2D reasoning by decomposing 2D understanding into complementary sub-tasks [7] Industry Implications - The launch of the DeepSeek-OCR2 model provides a technological upgrade direction for the OCR industry, enabling companies involved in graphic information processing and digital transformation services to optimize their products and expand business opportunities in finance, healthcare, and government sectors [8] - DeepSeek's commitment to an open-source technology route and the continuous release of high-performance model products will benefit developers and enterprises focusing on secondary development and deployment services [8] - The adaptation of DeepSeek's model on edge devices is pushing AI capabilities towards the edge, creating growth opportunities for companies involved in edge hardware development and edge computing solutions [8]

Group 1 - Northbound capital recorded a net buy of HKD 139.3 billion on January 21, with HK Stock Connect (Shanghai) contributing HKD 77.89 billion and HK Stock Connect (Shenzhen) contributing HKD 61.41 billion [1] - The most bought stocks included the Tracker Fund of Hong Kong (02800), Hang Seng China Enterprises (02828), and Alibaba-W (09988), while the most sold stocks were China Mobile (00941), Tencent (00700), and Hua Hong Semiconductor (01347) [1] - Alibaba-W received a net buy of HKD 10.77 billion, with significant growth in its AI model downloads, surpassing 100 million downloads [5] - China Life (02628) saw a net buy of HKD 10.07 billion, driven by positive expectations in the life insurance sector [5] - Kuaishou-W (01024) gained a net buy of HKD 7.33 billion, with its AI product showing a 350% increase in paid user growth [5] Group 2 - Semiconductor stocks showed divergence, with SMIC (00981) receiving a net buy of HKD 3.87 billion, while Hua Hong Semiconductor (01347) faced a net sell of HKD 1.2 billion [6] - TSMC's increased capital expenditure forecast indicates strong long-term demand driven by AI [6] - Shandong Gold (01787) received a net buy of HKD 3.03 billion, with gold prices hitting historical highs amid geopolitical tensions [7] - CNOOC (00883) saw a net buy of HKD 3.64 billion, with ongoing geopolitical tensions affecting oil prices [7] - Xiaomi Group-W (01810) and Meituan-W (03690) received net buys of HKD 5.89 billion and HKD 2.04 billion, respectively, while China Mobile (00941) and Tencent (00700) faced net sells of HKD 9.21 billion and HKD 2.71 billion [7]