每经编辑｜王瀚黎 王晓波 V国新办定于1月28日举行新闻发布会 2025年国资央企高质量发展情况 2 财政部、央行: 定于1月28日进行2( 国库现金管理商业银行定期存款（一期 3 美国2025年12月商品贸易帐公布 4 美国1月23日当周EIA汽油库存变动' 1 隔夜市场 美股三大指数收盘涨跌不一，道指跌0.83%，纳指涨0.91%，标普500指数涨0.41%；科技股普涨，英特尔涨超3%，亚马逊涨超2%。医疗保险股普跌，标普 1500管理型医疗指数下跌18%，创下自1997年以来的最大跌幅；联合健康跌超19%，CVS跌超14%。中概股涨跌不一，纳斯达克中国金龙指数涨0.48%，爱 奇艺、哔哩哔哩涨超3%，携程集团跌逾2%，小鹏汽车、理想汽车跌超1%。 国际油价大幅走高，截至发稿时，美油主力合约涨3.07%，报62.49美元/桶；布伦特原油主力合约涨2.9%，报66.65美元/桶。 国际贵金属延续强势，其中黄金再创历史新高，现货黄金盘中最高触及5187.96美元/盎司，截至发稿时，现货黄金涨3.33%，报5174.95美元/盎司；COMEX 黄金期货涨1.87%，报5177.4美元/盎司。现货白银涨8.11% ...

● 本报记者葛瑶 国际资本对中国市场的关注度显著回升。去年境外投资者对中国市场的低配幅度从约2.5%收窄至约 1.3%，减少了约1.2个百分点的低配。瑞银全球金融市场交易部主管贺立言(Neil Hosie)日前在接受中国 证券报记者专访时表示："如果外资进一步降低低配幅度，将意味着可观的潜在资金流入。" 贺立言表示，外资情绪正从此前的观望态度转向"谨慎乐观"。在他看来，中国市场估值优势凸显、企业 治理持续改善、科技创新动能不断增强，叠加全球投资者分散化配置需求上升，配置型资金的入场窗口 有望逐渐打开。瑞银预计中国股市2026年全年涨幅约18%，盈利增速约10%，机器人、人工智能、新能 源汽车、生物科技等科技创新赛道最受关注。 外资谨慎乐观 "谨慎但乐观"，贺立言用这一表述来概括当前外资的情绪基调。他观察到，与一年前同期相比，境外投 资者的心态已发生明显转变。"今年投资者正在做大量工作来深入了解中国公司，这和去年年初相比已 经有了很大不同。"他认为，这一变化在一定程度上受2025年A股市场表现带动：上证指数全年涨幅达 18.41%，沪深300指数涨幅超17%，强劲的市场表现开始吸引全球投资者关注。 交易型资金活 ...

● 本报记者 葛瑶 国际资本对中国市场的关注度显著回升。去年境外投资者对中国市场的低配幅度从约2.5%收窄至约 1.3%，减少了约1.2个百分点的低配。瑞银全球金融市场交易部主管贺立言（Neil Hosie）日前在接受中 国证券报记者专访时表示："如果外资进一步降低低配幅度，将意味着可观的潜在资金流入。" 关注科技创新投资主线 贺立言表示，外资情绪正从此前的观望态度转向"谨慎乐观"。在他看来，中国市场估值优势凸显、企业 治理持续改善、科技创新动能不断增强，叠加全球投资者分散化配置需求上升，配置型资金的入场窗口 有望逐渐打开。瑞银预计中国股市2026年全年涨幅约18%，盈利增速约10%，机器人、人工智能、新能 源汽车、生物科技等科技创新赛道最受关注。 外资谨慎乐观 "谨慎但乐观"，贺立言用这一表述来概括当前外资的情绪基调。他观察到，与一年前同期相比，境外投 资者的心态已发生明显转变。"今年投资者正在做大量工作来深入了解中国公司，这和去年年初相比已 经有了很大不同。"他认为，这一变化在一定程度上受2025年A股市场表现带动：上证指数全年涨幅达 18.41%，沪深300指数涨幅超17%，强劲的市场表现开始吸引全球投 ...

转自： InfoQ 刚刚，DeepSeek 发布了 新模型 DeepSeek-OCR 2，采用创新的DeepEncoder V2方法，让AI能够根据图像的含义动态重排图像的各个部分，更接近人类的视觉编码逻 辑。在具体实现上，DeepSeek 团队在论文中称采用了 Qwen2-0.5B 来实例化这一架构。 而在 DeepSeek-OCR 2 中，这一思路被进一步推进。 根据技术报告，DeepEncoder V2 不再将视觉编码视为一次静态的、固定策略的扫描过程，而是引入了语义驱动的动态编码机制。模型会在编码阶段就开始判断哪些 区域更可能承载关键信息，并据此调整视觉 token 的分配与表达方式。 换句话说，视觉编码不再只是"预处理"，而是已经提前进入了"理解阶段"。 和 DeepSeek 过往几乎所有重要发布一样，这一次依然选择了模型、代码与技术报告同时开源。项目、论文和模型权重已同步上线： 项目地址：https://github.com/deepseek-ai/DeepSeek-OCR-2 论文地址：https://github.com/deepseek-ai/DeepSeek-OCR-2/blob/main ...

Group 1 - The core argument is that the U.S. faces systemic deficiencies in power supply, which will hinder its development in artificial intelligence (AI), while China has a robust and sustainable power supply that supports AI evolution [1] - By mid-2025, the performance growth of individual chips will begin to plateau, shifting the focus from computing power to electricity supply, where China excels [1] - China's total installed power generation capacity is 3.8 billion kilowatts, with an estimated annual electricity consumption of approximately 10.4 trillion kilowatt-hours in 2025, more than double that of the U.S. [1] Group 2 - The U.S. AI development is likened to a mystical journey, while China's "AI+" initiative focuses on practical experience in specific sectors, fostering skilled professionals to enhance productivity and national competitiveness [2]

Group 1: Semiconductor Industry - The semiconductor industry is currently facing systemic cost pressures, with companies like Zhongwei Semiconductor and Guoke Micro announcing price increases of 15%-50% for products such as MCU and Norflash due to supply chain constraints and rising costs [1][2] - The packaging and testing costs have significantly increased, particularly for standard products like MCU and Norflash, leading to longer delivery cycles and greater operational pressure on companies [2] - The demand for DRAM and NAND Flash from major manufacturers is driving up orders for storage testing companies, with price increases of up to 30% expected as capacity utilization approaches full load [2] Group 2: Natural Gas Market - A severe winter storm has caused natural gas prices in the U.S. to surge, with Henry Hub futures reaching $7.43 per million British thermal units, a 140% increase since January 16 [3] - Spot prices in the northeastern U.S. have skyrocketed to between $50 and $150 per million British thermal units due to the storm, which has resulted in significant fatalities [3] - Short-term oil prices are under pressure from tariff policies and OPEC+ production increases, but geopolitical risks and global demand resilience may support oil price stability [3] Group 3: Artificial Intelligence - DeepSeek has launched a new document recognition model, DeepSeek-OCR2, which features a self-developed visual encoder that enhances semantic understanding [4] - The AI application landscape is expected to accelerate significantly by late 2025 to early 2026, with major tech companies making strategic moves in the AI space [4] - The upcoming release of DeepSeek's V4 model is anticipated to create competitive pressure on overseas closed model vendors, potentially reshaping the global AI market [5] Group 4: Brain-Computer Interface - The Brain-Computer Interface (BCI) industry is transitioning from laboratory research to clinical and consumer applications, with significant policy support and technological advancements driving this shift [5] - The BCI has been included in China's "14th Five-Year Plan" as a key development area, with independent billing projects set to be established by 2025 [5] - Non-invasive BCI technologies are already in mass production for rehabilitation and focus training, with companies like Neuralink planning to scale production of BCI devices [5] Group 5: Macro and Industry News - China Aviation Engine Corporation has outlined its strategic tasks for the "14th Five-Year Plan," focusing on accelerating the development of aviation engines and achieving technological self-reliance [6] - A major green hydrogen project in Inner Mongolia has undergone a significant capacity increase, now featuring 308 alkaline electrolysis cells and 152 PEM devices [7] - The National Energy Administration has established several standardization technical committees related to the hydrogen energy sector [7]

Core Insights - The article highlights significant advancements in AI technology with the release of new models and chips by various companies, indicating a competitive landscape in the AI sector. Group 1: AI Model Releases - DeepSeek has launched the DeepSeek-OCR 2 model, enhancing optical character recognition capabilities [1] - Alibaba has officially released its flagship reasoning model, Qwen3-Max-Thinking, marking a significant step in AI development [1] - Kimi announced the release and open-sourcing of the Kimi K2.5 model, contributing to the open-source AI community [1] - MiniMax introduced the MiniMax M2.1 × Clawdbot, aimed at creating an open-source AI assistant to build superintelligent workflows [1] Group 2: AI Hardware Developments - Microsoft has unveiled the next-generation AI chip, Maia 200, which is expected to improve AI processing capabilities [1] - NVIDIA has released the AI "Earth-2" system, designed to enhance the accuracy of weather forecasting [1]

Core Viewpoint - The article highlights the recent advancements in domestic AI models, particularly focusing on the release of new models by DeepSeek, Kimi, and Alibaba, which demonstrate significant improvements in performance and capabilities in visual understanding and reasoning [2][8][9]. Group 1: DeepSeek's Innovations - DeepSeek released the new DeepSeek-OCR 2 model on January 27, utilizing the innovative DeepEncoder V2 method, allowing AI to dynamically rearrange image components based on their meaning, mimicking human visual encoding logic [2][3]. - The DeepSeek-OCR 2 model achieved a score of 91.09% on the OmniDocBench v1.5 benchmark, representing a 3.73% improvement over its predecessor [3][4]. - The model maintains high precision while controlling computational costs, with visual token counts limited between 256 and 1120, aligning with Google’s Gemini-3 Pro [3][4]. Group 2: Kimi's Model Release - Kimi launched the Kimi K2.5 model, which automatically updated the previous K2 model without user intervention, aimed at enhancing response speed, reasoning ability, and multi-turn dialogue stability [8]. - Kimi K2.5 achieved top scores in various agent evaluations, including HLE, BrowseComp, and DeepSearchQA, marking it as Kimi's most intelligent model to date [8]. Group 3: Alibaba's Qwen3-Max-Thinking - Alibaba introduced the Qwen3-Max-Thinking model, which surpassed leading models like GPT-5.2 and Claude Opus 4.5 in multiple key performance benchmarks, setting new global records [8][9]. - The new model features over one trillion parameters and incorporates a novel test-time scaling mechanism, significantly enhancing reasoning performance while being more economical [9]. - Qwen3-Max-Thinking also improves the model's ability to autonomously use tools, reducing hallucinations and laying the groundwork for solving complex real-world tasks [9].

"世界不是由事实构成的，而是由事实之间的关系构成的。" 如果用维特根斯坦的理论来反推 AI 的能力边界，智能的上限或许从一开始就不取决于模型"知道多少"，而取决于它是否理解Context(语境)、规则，以及 这些知识在不同场景中如何被使用。 也正是在这一意义上，今天更有竞争力的模型们，开始逐步逼近维特根斯坦后期所说的"语言游戏"：意义并不来自词本身，而来自使用。能否参与这种游 戏，决定了 AI 只是一个高效的工具，还是正在进入更深层的认知结构。 但这一变化，并没有被舆论第一时间捕捉。过去两年，舆论场被 ChatGPT 与 Claude 轮番占据，行业习惯将 AI 视为一个整体的、线性的竞赛，更强的模 型、更大的参数、更通用的智能被视为唯一的进化方向。 身处一线的从业者，先于市场感知到了"温差"。前OpenAI成员姚顺雨，便在此前AGI会谈上分享到： AI 在 To C 端和 To B 端正遵循不同的发展轨迹。 从 GPT-4 到后续迭代版本，普通 C 端用户的体感差异微乎其微；但另一边，Claude，已开始深入编程等核心环节，改变程序员们的工作模式。 陷入马太效应的"垂直整合" 过去两年，市场曾笃信"模型+应 ...

Core Insights - Tencent has acknowledged its slow progress in AI compared to competitors like ByteDance and Alibaba, estimating a lag of 9 months to 1 year in AI development [2][3] - The company is focusing on integrating AI with its existing platforms and is undergoing organizational adjustments to enhance its AI capabilities [2][4] Group 1: AI Strategy and Market Position - Tencent's leadership has identified four key companies in the AI sector, with Tencent being the last to enter the race, thus in a catch-up phase [3] - The company recognizes that its AI strategy needs to adapt to how different industries are being transformed by AI, particularly in search and programming [4][5] - Tencent's AI initiatives are seen as an accelerator for existing business, with AI-driven advertising revenue showing significant growth [13][15] Group 2: Challenges and Internal Reflections - Internal reflections reveal that Tencent's focus on product engineering has led to a lack of comprehensive AI research, which is critical for long-term success [8] - The company has faced issues with its foundational AI models, leading to ineffective product development and low user retention rates [11][12] - Despite recognizing AI as a key investment area, Tencent's capital expenditure on AI has not significantly increased, indicating potential resource allocation challenges [11] Group 3: Future Directions and Innovations - Tencent is restructuring its AI research teams and has established new departments to enhance its AI infrastructure and capabilities [15][16] - The integration of AI into WeChat is seen as a future opportunity, with the potential for a unique AI agent that aligns with user needs and company values [16] - The company aims to leverage its existing ecosystem to create innovative AI applications, enhancing user interaction and efficiency [16]