Core Insights - The release of DeepSeek-V3.2 and its enhanced version V3.2-Speciale marks a significant advancement in open-source large models, achieving top-tier performance and practicality, particularly in reasoning capabilities and tool integration [2][3]. Group 1: Performance and Innovation - DeepSeek-V3.2 series has reached a breakthrough in core reasoning capabilities, matching the performance of top closed-source models and significantly outperforming some open-source models focused on long contexts [2]. - The Speciale version has excelled in international competitions, achieving gold medals in events like the International Mathematical Olympiad (IMO) and the International Collegiate Programming Contest (ICPC), where it ranked second among human competitors [2]. - The model innovatively integrates thinking modes with tool invocation, enhancing the agent's generalization and execution capabilities in complex scenarios [3]. Group 2: Technical Advancements - DeepSeek-V3.2 is the first open-source model to systematically incorporate chain-of-thought reasoning into the tool invocation process, utilizing a unique large-scale agent training data synthesis method [3]. - The model has undergone reinforcement learning across over 85,000 complex instructions in more than 1,800 environments, achieving the highest level among open-source models in untrained tool invocation assessments [3]. Group 3: Ecosystem and Market Impact - The comprehensive upgrade of DeepSeek-V3.2's open-source and API services is expected to accelerate technological penetration and drive a transformation in industrial application paradigms [4]. - The open strategy, combining performance and ecosystem openness, significantly lowers the application barriers for enterprises and developers, potentially leading to a large-scale, practical deployment of open-source models [4]. - This approach is anticipated to attract numerous developers to build vertical applications based on DeepSeek, forming a robust open-source application ecosystem centered around it [4].

Core Viewpoint - Fluence is negotiating over 30 GWh of AIDC (Artificial Intelligence Data Center) energy storage, which is seen as a significant emerging market opportunity as the demand for energy in data centers is expected to rise sharply in the coming years [2][3]. Group 1: AIDC Energy Storage Market - Fluence is in discussions for over 30 GWh of energy storage projects, with 80% of these projects initiated after the end of Q4 2025, indicating a growing market for data center energy storage solutions [2]. - The energy consumption of data centers in the U.S. reached 176 TWh in 2023, accounting for 4.4% of the total electricity consumption, with projections indicating a growth rate of 13%-27% annually from 2023 to 2028, potentially increasing consumption to between 325-580 TWh by 2028 [2]. - If 50 GW of new data center capacity is added by 2030, the electricity gap in the U.S. could reach 23 GW, and this gap would be larger when considering the retirement of existing power plants [2]. Group 2: Short-term and Long-term Solutions - In the short term, energy storage can help data centers manage peak loads and frequency regulation, which is crucial given the aging U.S. power grid and its limited ability to adjust [3]. - The integration of energy storage is expected to facilitate the connection of data centers to the grid, with current connection times ranging from 3 to 7 years depending on the location [3]. - Long-term, solar and storage solutions may evolve into primary power sources for data centers, as the economic viability of solar storage is becoming evident compared to traditional gas turbine solutions, which have longer supply cycles [3].

Core Viewpoint - The development of AIDC may exacerbate electricity shortages in the U.S., with data center energy storage serving as a potential solution. Fluence is negotiating over 30 GWh of AIDC energy storage projects, indicating a significant emerging market opportunity [2][3]. Group 1: AIDC Development and Energy Demand - AIDC's high energy consumption could lead to increased electricity shortages in the U.S. According to the DOE, U.S. data centers consumed 176 TWh in 2023, accounting for 4.4% of total electricity consumption. The demand for electricity from data centers is expected to grow annually by 13%-27% from 2023 to 2028, potentially reaching 325-580 TWh by 2028, which would increase their share of total U.S. electricity consumption to 6.7%-12% [2][3]. - If 50 GW of new data centers are added by 2030, the projected electricity gap in the U.S. could reach 23 GW, and this gap would be larger when considering the retirement of existing power plants [2]. Group 2: Short-term and Long-term Solutions - In the short term, energy storage can help data centers manage peak load and frequency regulation, which is crucial given the aging U.S. power grid. The interconnection process for data centers can take several years, with estimates of about 3 years in Chicago and 7 years in Virginia [3]. - Long-term, solar and storage solutions may evolve into self-sufficient power sources for data centers. Currently, gas turbines are the mainstream solution, but their supply chain can take over 3 years. In contrast, solar storage has already demonstrated economic viability and offers advantages in interconnection timelines compared to gas turbines [3].

Core Insights - The article focuses on optimizing a timing model based on price and volume factors, addressing issues of instability, signal volatility, and the reliability of single signals [1][2]. Factor Selection - The model employs a dual standard of group IC and threshold settings to tackle the challenge of unstable effectiveness, ensuring that selected factors can consistently predict outcomes across different value ranges [2]. Model Training and Signal Generation - A strategy of random grouping and independent training is used to filter noise and balance signal robustness. The signal generation process involves rolling smoothing and multi-group voting to ensure accurate and stable timing signals [3]. - Backtesting from 2019 to September 2025 shows significant improvements over benchmarks, with a 1-day signal yielding an annualized return of 3.61% and a Sharpe ratio of 1.12, outperforming the benchmark [3].

报告导读： 豆包推出 AI 手机助手预览版，持续探索 AI 大模型对手机的赋能，推动 Agent 形态的形成。 AI 手机的时代即将到来， 同时 更加丰富的端侧硬件生态加快崛起 投资建议。 豆包推出 AI 手机助手预览版，持续探索 AI 大模型对手机的赋能，推动 Agent 形态的形成。 AI 手机的时代即将到来，同时更加丰富的端侧硬件 生态加快崛起。 豆包发布 AI 手机助手预览版，作为系统级 Agent 可以执行复杂指令，实现从信息搜索到信息处理，再到跨应用沟通的完整闭环。 豆包手机助手实现了与操 作系统的深度融合，比单一 APP 有更高系统权限，可以调用底层硬件能力和系统服务。 用户无需点击图标进入，而是通过侧边栏、实体键或语音直接唤醒， 随时随地覆盖在任何界面之上。同时豆包手机助手运行不打断当前正在进行的 app 进程。 豆包助手能够"看懂"用户当前屏幕上的内容，并基于此进行对话或 操作（比如生成回复、协助用户处理微信好友的请求）。 同时豆包助手能够识别屏幕上的 UI 元素，模拟人手进行点击、滑动和输入等操作，实现复杂的跨应 用任务；比如帮助用户在多个购物平台上比价某商品并下单；出国做旅行攻略，在小红 ...

上游资源：煤价环比回落，工业金属价格上涨。 1 ）煤炭： 煤价环比 -2.2% ，尽管供给仍偏紧，但需求端短期难有明显超预期空间，煤价回吐部分涨幅 ； 2 ）有色： 降息预期升温，工业金属价格上涨。 报告导读： 中观景气延续分化的增长格局，新兴科技景气仍强，高性能存储价格延续快速 上涨，游戏供给偏宽松；服务消费景气明显提升，地产周期和耐用品需求仍承压。 服务消费景气提升，科技硬件延续涨价。 上周（ 11.24-11.30 ）中观景气表现分化，值得关注： 1 ）内需景气线索有所增多，冰雪出行和电影市场景气度 显著提升，或反映地产和耐用品消费收缩之余，"吃喝玩乐"相关的服务型和大众品消费复苏趋势持续显现。 2 ）新兴科技行业延续高景气，但短期 AI 泡沫叙 事影响下， TMT 硬件景气增长的持续性有赖于 AI 应用取得积极进展。后续重点关注 AI 应用商业化进展。 3 ）建工需求偏弱，内需资源品大多偏弱震荡， 海外降息预期再度升温，国际金属价格大幅上涨；受铁矿增产影响，干散运价格环比延续提升。 下游消费：服务消费景气显著提升，地产耐用品仍承压。 1 ）服务消费： 国内冰雪游景气度显著提升，根据同程旅行，广州 - ...

报告来源 以上内容节选自国泰海通证券已发布的证券研究报告。 报告名称： Gemini 3、TPU、端侧AI应用更新报告——模型多模态升级加速端侧AI落地，TPU冲击算力 格局；报告日期：2025.12.02 报告作者： 秦和平（分析师），登记编号：S0880523110003 刁云鹏（研究助理），登记编号：S0880125070016 重要提醒 本订阅号所载内容仅面向国泰海通证券研究服务签约客户。因本资料暂时无法设置访问限制，根据《证 券期货投资者适当性管理办法》的要求，若您并非国泰海通证券研究服务签约客户，为保证服务质量、 控制投资风险，还请取消关注，请勿订阅、接收或使用本订阅号中的任何信息。我们对由此给您造成的 不便表示诚挚歉意，非常感谢您的理解与配合！如有任何疑问，敬请按照文末联系方式与我们联系。 报告导读： 模型：预训练 Scaling Law 仍然成立；算力： TPU 助谷歌构建全栈 AI 生 态，长期或与英伟达 GPU 互补；应用：多模态推理能力为端侧 GUI 操控提供可能，豆包 手机助手率先落地，看好谷歌全栈集成、苹果系统掌控、阿里模型能力。 Gemini 验证了预训练 Scaling Law ...

Core Viewpoint - The Q4 style rotation model indicates signals for small-cap and growth stocks, with a focus on sectors such as electric equipment and renewable energy for December [1][2]. Style Rotation Model - The Q4 style rotation model has issued signals favoring small-cap stocks, with a comprehensive score of -1 for the dual-driven rotation strategy as of September 30, 2025 [3]. - The value-growth style rotation model shows a comprehensive score of -3 for the dual-driven rotation strategy, indicating a preference for growth stocks [4]. Industry Rotation Insights - In November, the composite factor strategy yielded an excess return of -0.58%, while the single-factor long strategy had an excess return of -0.83% [4]. - For December, the single-factor long strategy recommends bullish sectors including banking, construction, non-bank financials, and electric equipment and renewable energy. The composite factor strategy suggests bullish sectors such as telecommunications, comprehensive finance, computer technology, electric equipment and renewable energy, and utilities [4].

Group 1: Style Rotation Insights - The Q4 style rotation model indicates signals for small-cap and growth stocks [2][3] - The dual-driven rotation strategy for Q4 has a composite score of -1, predicting a focus on small-cap stocks [3] - The value-growth style rotation model shows a composite score of -3, suggesting a preference for growth stocks [4] Group 2: Industry Rotation Analysis - In November, the composite factor strategy yielded an excess return of -0.58%, while the single-factor long strategy had an excess return of -0.83% [4] - For December, the recommended long industries based on single-factor strategies include banking, construction, non-bank financials, and electric equipment & new energy [4] - The composite factor strategy recommends long positions in telecommunications, comprehensive finance, computers, electric equipment & new energy, and utilities [4] Group 3: Pharmaceutical Sector Performance - In November 2025, the pharmaceutical sector underperformed the broader market, with the SW pharmaceutical and biological index declining by 3.6% compared to a 1.7% drop in the Shanghai Composite Index [7] - The relative premium level of the pharmaceutical sector is currently at 72.6%, indicating a normal valuation level compared to all A-shares [7] - In the Hong Kong market, the pharmaceutical sector performed similarly to the market, with the Hang Seng Medical Care index at -0.1% and the biotechnology sector at +0.4% [7] Group 4: U.S. Pharmaceutical Market Trends - In November 2025, the U.S. pharmaceutical sector outperformed the broader market, with the S&P Healthcare Select Sector Index rising by 9.1% compared to a 0.1% increase in the S&P 500 [8] - Notable gainers in the S&P 500 healthcare component included Eli Lilly (+25%) and Solventum (+23%) [8]

全球主要债市收益率普遍下行，美债呈现长端降幅大于短端的牛陡特征， 30 年期降 5.2 个基点，英债 10 年期大跌 9.34 个基点领跑发达市场。 德债温和 下行反映欧央行谨慎立场，日债短端上行显示央行政策正常化压力。信用利差显著压缩，投资级企业债整体降 11 个基点，高收益债降 29 个基点至 6.58% ， VIX 暴跌 30.2% 至 16.35 。 报告导读： 在不确定性窗口期以中久期高评级策略平衡收益获取与风险控制。 全球债市聚焦欧洲财政风险、美国数据真空与新兴市场信用改善三大主线。 欧洲央行警告主权债供给压力加剧且央行购债规模萎缩，利率风险上升。美国政 府停摆导致关键经济数据永久缺失， 12 月降息概率从 95% 降至 50% ，市场陷入政策不确定性。新兴市场凭借十年最佳评级周期和财政改革吸引资金流 入，固定收益配置亮点凸显。 点心债发行 41 只、规模 953.83 亿元，央行发行 450 亿元央票占比 47.2% 成为最大亮点，金融债占比 86% 。 中资美元 / 港币债发行 7 只、规模 14.786 亿美元 +1.5 亿港币。整体发行结构以银行金融债为主导，城投债票面利率集中在 5-7% ...