Core Insights - DeepSeek has released a new paper focusing on the conditional memory module of large models, suggesting it will be a core modeling primitive in the next generation of sparse large models [2][5][7] Group 1: Research and Development - The new paper, co-authored with Peking University, is titled "Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models" [5] - The research identifies two distinct tasks within large models: deep dynamic computation for combinatorial reasoning and static knowledge retrieval, highlighting inefficiencies in the current Transformer architecture [5][6] - DeepSeek introduces conditional memory as a supplementary sparse dimension to optimize the balance between neural computation (MoE) and static memory (Engram) [6][7] Group 2: Performance and Implications - The team discovered a U-shaped scaling law indicating that the mixed sparse capacity allocation between MoE experts and Engram memory significantly outperforms pure MoE baseline models [6] - The introduction of the memory module not only aids knowledge retrieval but also shows significant improvements in general reasoning, coding, and mathematical tasks [6][7] - The paper essentially proposes a "division of labor" optimization for large models, allowing specialized modules to handle specific tasks more efficiently [6][7] Group 3: Future Developments - Industry speculation suggests that the proposed conditional memory may be part of the technical architecture for DeepSeek's upcoming flagship model, DeepSeek V4, expected to be released around February [7] - Initial tests indicate that V4 may surpass other leading models in programming capabilities, with the previous V3 model having already outperformed OpenAI's GPT-5 and Google's Gemini 3.0 Pro in various benchmarks [7]

Core Insights - DeepSeek has released a new paper focusing on the conditional memory module of large models, suggesting it will be a core modeling primitive in the next generation of sparse large models [1][4]. Group 1: Research Findings - The new paper, co-authored with Peking University, is titled "Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models" and highlights the need for a native knowledge retrieval mechanism in existing Transformer architectures [4]. - The research identifies two distinct tasks in large models: deep dynamic computation for combinatorial reasoning and static knowledge retrieval, indicating that current models inefficiently simulate retrieval processes [4][5]. - DeepSeek introduces conditional memory as a supplementary dimension of sparsity, optimizing the trade-off between mixture of experts (MoE) and static memory (Engram) [4][6]. Group 2: Performance Improvements - The team discovered a U-shaped scaling law, showing that the mixed sparse capacity allocation between MoE experts and Engram memory significantly outperforms pure MoE baseline models [5]. - The introduction of the memory module not only aids knowledge retrieval but also yields notable improvements in general reasoning, coding, and mathematical tasks [5][6]. - The paper essentially proposes a "division of labor" optimization for large models, allowing specialized modules to handle specific tasks, thereby enhancing efficiency and resource allocation [6]. Group 3: Future Developments - Industry speculation suggests that the proposed conditional memory may be integral to the architecture of DeepSeek's upcoming flagship model, DeepSeek V4, expected to be released around February [6]. - Initial tests indicate that V4 may surpass other leading models in programming capabilities, with the previous model, V3, having already outperformed OpenAI's GPT-5 and Google's Gemini 3.0 Pro in various benchmarks [6].

2026.01.13 使用人工智能模型来设计和发现新药物，是目前全球制药巨头加速布局的方向，此举目标是缩短新药研 发上市所需的时间。 本文字数：1275，阅读时长大约2分钟 作者 |第一财经 钱童心 当地时间1月12日，摩根大通医疗健康大会（JPM Conference）在旧金山正式召开。大会首日，全球市 值最大的科技公司英伟达与全球市值最大的制药公司礼来宣布，将在五年内斥资10亿美元在旧金山湾区 建立一个新的联合研究实验室，以加速AI药物研发进程。 截至发稿，英伟达市值为4.5万亿美元，位居全球市值最大公司榜首；礼来市值稳定在1万亿美元上方， 牢牢占据全球市值最大制药企业的地位。 据介绍，该实验室将使用英伟达最新一代AI芯片Vera Rubin。就在几个月前，礼来曾表示，正使用1000 多个英伟达当前一代AI芯片Grace Blackwell构建一台超级计算机。这也有望成为全球最强大的AI计算机 之一。 微信编辑| 七三 第一财经持续追踪财经热点。若您掌握公司动态、行业趋势、金融事件等有价值的线索，欢迎提供。专 用邮箱：bianjibu@yicai.com （注：我们会对线索进行核实。您的隐私将严格保密。） ...

2026.01.13 本文字数：518，阅读时长大约1分钟 作者 |第一财经 栾立 第一财经持续追踪财经热点。若您掌握公司动态、行业趋势、金融事件等有价值的线索，欢迎提供。专 用邮箱：bianjibu@yicai.com （注：我们会对线索进行核实。您的隐私将严格保密。） i茅台"秒空"仍在继续，今日（1月13日）标价4199元的53%vol 500ml陈年贵州茅台酒（15）首日在i茅台 app上线也告秒空。 官方消息显示，2026年1月13日起，陈年贵州茅台酒（15）也正式登录i茅台APP抢购，陈年贵州茅台酒 （15）也是茅台的"塔尖"产品。记者注意到，9点09分刚到，陈年贵州茅台酒（15）已经被抢购一空。 据酒商介绍，陈年贵州茅台酒（15）平时多用于礼赠，需求并不是特别多。其也属于2025年市场价格倒 挂比较严重的产品，近期批发价一般在3900元/瓶左右。尚不清楚i茅台每日的陈年贵州茅台酒（15）供 给量是多少，因此并不好做判断，而且也不排除有个人投机者在其中操作。 根据独立白酒评论员肖竹青介绍，近期茅台刚刚调整过陈年贵州茅台酒（15）的价格，出厂价从5399 元/瓶下调至3409元/瓶，零售价从5999 ...

| 智明达 | 53.90 | -10.17% | | --- | --- | --- | | 588988989 | | | | 航天电子 | 28.40 | -10.01% | | 600879.SH | | | | 中国卫通 | 48.49 | -10.00% | | 601698.SH | | | | 航天长峰 | 26.45 | -10.00% | | 600855.SH | | | | 航天发展 | 36.00 | -10.00% | | 000547.SZ | | | | 航天动力 | 44.64 | -10.00% | | 600343.SH | | | | 中天火箭 | 88.83 | -10.00% | | 003009.SZ | | | | 震刺示 | 60.21 | -10.00% | | 002151.SZ | | | | 航天科技 | 31.50 | -10.00% | | 000901.SZ | | | | 航天电器 | 63.67 | -9.99% | | 002025.SZ | | | | 航天晨光 | 34.69 | -9.99% | | 600501 SH | | | I ...

当地时间1月12日，苹果与谷歌宣布达成一项多年期战略合作协议。根据双方披露的信息，谷歌的 Gemini核心模型架构将被用于支持下一代Apple Foundation Models，并成为Siri新一轮升级的底层技术基 础。 在一份声明中，苹果将与谷歌的这次合作描述为Apple Foundation Models提供了"强大的基础"。 谷歌方面则表示，"苹果与谷歌已进入多年合作阶段，下一代苹果基础模型将基于谷歌的Gemini模型和 云技术。这些模型将助力未来Apple Intelligence功能的实现，包括预计今年内推出的、更具个性化的 Siri。" 双方的合作细节并未公开。但有消息称，苹果预计每年将向谷歌支付约10亿美元的技术许可费用。 这意味着，长期坚持自研路线的苹果，在生成式人工智能的"核心引擎"层面，首次正式引入来自最大竞 争对手之一的基础模型技术。 消息公布数小时后，特斯拉与xAI创始人埃隆·马斯克在X平台连续发文，对这一合作表达强烈不满，称 其将进一步加剧科技行业的"权力集中"。 这一表态并非孤立。马斯克旗下的xAI过去一年里已多次公开批评当前的AI产业结构，并曾对苹果和 OpenAI提起诉讼，指 ...

09:25 A股开盘丨三大指数集体高开 沪指高开0.11%，深成指高开0.21%，创业板指高开0.07%，科创综指低开0.12%。 | 代码 | 名称 | 两日图 | 现价 | 涨跌 | 涨跌幅 | | --- | --- | --- | --- | --- | --- | | 000001 | 上证指数 | W/ | 4169.70 | 4.42 | 0.11% | | 399001 | 深证成指 | NA | 14397.68 | 30.78 | 0.21% | | 399006 | 创业板指 | 3 | 3390.69 | 2.35 | 0.07% | | 000680 | 科创综指 | No | 1853.18 | -2.21 | -0.12% | 2026.01.13 本文字数：596，阅读时长大约1分钟 作者 |一财阿驴 09:30AI应用方向延续强势，引力传媒7天6板，易点天下、卓易信息、迪安诊断、利欧股份、省广集团 均3连板。 恒生指数高开1.32%，恒生科技指数涨1.93%。科网股全线走强，阿里巴巴、阿里健康涨近5%，小鹏汽 车涨逾4%，金蝶国际、比亚迪股份涨逾3%。兆易创新H股上市高开45 ...

2026.01.13 本文字数：2660，阅读时长大约4分钟 作者 |第一财经 安卓 为214.27倍，最新市净率为5.67倍。且该公司2024年营业收入同比减少25.06%、扣非后归母净利润同比 减少96.67%，盈利水平相对有限。 部分上市公司仍处于亏损状态，比如，理工导航自2026年1月7日以来收盘价格累计上涨幅度为 55.45%，该公司2024年扣非后净利润为亏损877.50万元，2025年1-9月扣非后净利润为亏损716.52万元。 自2025年12月份以来，商业航天概念股一改过往的起伏波动，呈现出强劲的增长态势，成为岁末年初资 本市场"最靓的仔"。 以万得商业航天主题指数为例，2025年12月1日至2026年1月12日期间，该指数累计涨幅达89%，大幅跑 赢同期沪深300指数与一众万得热门指数，高居涨幅榜榜首。 在个股方面，集AI、新材料、商业航天、机器人、量子科技等热门概念于一身的志特新材 （300986.SZ）股价已连续6个交易日"20CM"涨停，年内涨幅达198.57%，成为2026年首个翻倍股。 1月12日晚间，志特新材发布股价异动暨停牌核查公告，该公司最新滚动市盈率达到了104 倍，与所 ...

1月13日，香港恒生指数开盘涨1.32%，恒生科技指数涨近2%。 | 代码 | 名称 | 现价 涨跌 | 涨跌幅 | | --- | --- | --- | --- | | H2I | 恒生指数 | 26958.52c 350.04 | 1.32% | | HSTECH | 恒生科技 | 5976.22c 113.02 | 1.93% | | HSBIO | 恒生生物科技 | 16304.67c 129.49 | 0.80% | | HSCEI | 恒生中国企业指数 | 9335.53c 115.45 | 1.25% | | HSCI | 恒生综合指数 | 4141.07c 56.72 | 1.39% | 科网股大涨，阿里巴巴、阿里健康涨近5%，小鹏汽车涨逾4%，金蝶国际、比亚迪股份涨逾3%，腾讯控 股、京东集团、哔哩哔哩、联想集团涨超2%，百度集团涨近2%。 | < w | 恒生科技(HSTECH) | | Q | | --- | --- | --- | --- | | 01-13 09:20:46 资讯 成分 资料 | | | 相关基金 月度收 | | 名称 | | 现价 | 涨跌 三 | | 阿里巴巴- ...

Group 1 - The U.S. Labor Department is set to release the Consumer Price Index (CPI) report for December, with expectations of a slight increase in inflation, remaining above the Federal Reserve's 2% target due to data collection disruptions from a government shutdown [2][3] - Analysts predict a 0.3% month-on-month increase in overall inflation for December, with a year-on-year increase of 2.7%. Core inflation, excluding volatile food and energy prices, is expected to rise by 0.26% month-on-month and 2.7% year-on-year, both higher than November's actual readings [3][4] - The impact of tariffs is expected to continue pushing up prices, particularly in food, clothing, and automotive sectors, despite some tariffs being lifted at the end of last year [3][4] Group 2 - The market's expectations for Federal Reserve rate cuts have diminished, with the probability of a 25 basis point cut in the upcoming meetings dropping significantly, particularly from 79% to 40% for the April meeting [2][5] - The December non-farm payroll report indicated a slower job growth rate, which has led to a more stable labor market, but concerns about the job market have increased among households, raising short-term inflation expectations [5][6] - The potential political interference from the Trump administration regarding Federal Reserve Chairman Powell may deter the Fed from cutting rates, as officials aim to avoid perceptions of yielding to political pressure [6]