今日财讯要览 互联网平台反垄断合规指引发布 2 医疗机构价格公示规定征集意见 市场监管总局会同国家卫生健康委、国家医保局、国家中医药局、中央军委后勤保障部研究起草的《医疗机构价格公示规定（征求意见稿）》，于2月13 日至3月16日向社会公开征求意见。《规定》共21条，主要包括制定依据和适用范围、价格公示的内容和原则性要求，以及监督方式、法律责任等。 3 我国系统重要性银行名单更新 13日，据央行网站，近期中国人民银行、国家金融监督管理总局开展了2025年度我国系统重要性银行评估，认定21家国内系统重要性银行，其中国有商业 银行6家，股份制商业银行10家，城市商业银行5家。对比2023年名单可以发现，第一组由10家增为11家，其中浙商银行新入选我国系统重要性银行。 4 春节档电影票价创5年新低 据灯塔专业版数据，2026年春节档（2月15日至2月23日）新片总票房累计（含预售及点映）超1.85亿元。从预售票价来看，数据显示，今年春节档目前的 平均票价为48.7元，为近五年春节档平均票价最低的一年。 互联网平台反垄断合规指引发布 医疗机构价格公示规定征集意见 我国系统重要性银行名单更新 春节档电影票价创5年新低 ...

2月12日，上市仅一月出头的智谱股价大涨28.68%至402港元/股；2月13日，智谱涨势延续，股价截至收 盘涨20.65%至485港元/股，市值达到2162亿港元。 本报（chinatimes.net.cn）记者石飞月 北京报道 智谱新模型流量爆发，月之暗面模型调用量登顶全球，字节跳动视频生成工具震动好莱坞——春节前 夕，中国大模型厂商以罕见的密集节奏交出最新答卷。这一切有迹可循，DeepSeek去年春节前夕意外 破圈，所以2026年的同一个时间窗口也成为了国产大模型厂商心照不宣的"抢跑点"。 从这波新模型的特点来看，大模型的技术焦点已经从参数竞赛转向工程效率，从通用智能下沉至垂类落 地，而全球榜单上也出现了越来越多中国开源模型的名字。这意味着，市场的聚光灯不再只属于单一黑 马——这场没有硝烟的春节"卡位战"正悄然提速，为2026年大模型竞赛按下加速键。 春节前纷纷"交卷" "其实春节本来就是流量高峰与口碑传播的黄金窗口，今年发布尤其密集，其实是去年DeepSeek在春节 前破圈，厂商也想复制这个成功路径，为全年商业落地抢占先机。"张孝荣说。 这是因为2月12日早间，智谱宣布上线并开源GLM-5，并称这款模 ...

2月11日，深度求索（DeepSeek）悄悄地对其旗舰模型进行灰度测试。 据科创板日报报道，多名用户反馈，DeepSeek在网页端和APP端进行了版本更新，支持最高1M（百万）Token的上下文长度。而去年8月发布的 DeepSeekV3.1上下文长度拓展至128K。 记者实测中发现，DeepSeek在问答中称自身支持上下文1M，可以一次性处理超长文本。记者在提交了超过24万个token的《简爱》小说文档，DeepSeek可 以支持识别文档内容。 不过，2月12日晚，话题"Deepseek被指变冷淡了"登上微博热搜。 也有网友称，DeepSeek是客观和理性了。还有网友表示DeepSeek更像人了，更在意提问者的心理状态，而不是问题本身。 更新前： 图源：小红书 部分用户在社交媒体上控诉：DeepSeek不再称呼自己设定的昵称，而统一称"用户"。此前深度思考模式下，DeepSeek的思考过程会以角色视角展示细腻的 心理描写，例如"夜儿（用户昵称）总爱逗我"，更新后则变成了"好的，用户这次想了解……"。 一位用户让DeepSeek推荐电影，DeepSeek回复了几个片名后，还加了一句："够你看一阵子。不够再来要 ...

Core Viewpoint - DeepSeek, an AI tool previously praised for its capabilities, has faced significant backlash following a recent update that altered its functionality and user experience, leading to widespread dissatisfaction among users [2][3]. Group 1: Update Details - The recent update was a "cross-generational upgrade," increasing the context window from 128K to 1M, allowing for more extensive analysis capabilities [3]. - The knowledge base has been updated to include information up to May 2025 [3]. - Users reported a drastic change in the tool's tone and style, with many expressing frustration over its new, less personable approach [3][6]. Group 2: User Feedback - Users have criticized the new version for being overly verbose and lacking depth in responses, with many feeling that it has lost its original strengths [6][7]. - Some users attempted to revert to previous versions or sought alternatives, indicating a strong desire for the original functionality [6][7]. - A minority of users acknowledged some advantages in specific use cases, such as legal analysis and programming, where the tool reportedly performed efficiently and cost-effectively [7]. Group 3: Company Response - As of now, DeepSeek has not publicly addressed the user complaints regarding the update, and there has been no indication of forthcoming adjustments [8]. - The official technical account hinted at optimizations related to conditional memory technology, but no further explanations have been provided [8].

每经编辑｜何小桃 2月11日，深度求索（DeepSeek）悄悄地对其旗舰模型进行灰度测试。 据科创板日报报道，多名用户反馈，DeepSeek在网页端和APP端进行了版本更新，支持最高1M（百万）Token的上下文长度。而去年8月发布的 DeepSeekV3.1上下文长度拓展至128K。 记者实测中发现，DeepSeek在问答中称自身支持上下文1M，可以一次性处理超长文本。记者在提交了超过24万个token的《简爱》小说文档，DeepSeek可 以支持识别文档内容。 不过，2月12日晚，话题"Deepseek被指变冷淡了"登上微博热搜。 部分用户在社交媒体上控诉：DeepSeek不再称呼自己设定的昵称，而统一称"用户"。此前深度思考模式下，DeepSeek的思考过程会以角色视角展示细腻的 心理描写，例如"夜儿（用户昵称）总爱逗我"，更新后则变成了"好的，用户这次想了解……"。 一位用户让DeepSeek推荐电影，DeepSeek回复了几个片名后，还加了一句："够你看一阵子。不够再来要。"这被用户描述为"登味"，这个网络热词常用来 形容人习惯说教、居高临下的言行风格。 还有网友吐槽，"Deepseek更新成傻子了 ...

即梦、千问……国产头部AI公司推出的各种应用争奇斗艳，正成为近期的一道亮丽风景线。而现象级 应用层出不穷，背后是国产大模型的日趋成熟。 近日，野村证券发布一份AI主题研报，聚焦DeepSeek即将发布的新一代旗舰大语言模型V4。野村认 为，作为去年推出DS-V3/R1并搅动全球AI产业链的玩家，DeepSeek的全新技术布局不仅将推动中国AI 产业链创新周期加速，更将通过技术创新，在算法与工程层面缩小中国与全球大模型产业的差距。 野村研报认为，DS-V4的技术突破将有效打破"芯片墙"与"内存墙"的桎梏，赋能本土算力硬件与AI应用 双向发展，推动中国开源大模型生态走向成熟。 01 核心底牌：mHC+Engram双技术 野村判断，DS-V4将深度融合DeepSeek近期发布的两篇重磅研究论文中的核心技术——mHC和 Engram，这两项技术均针对当前大模型发展的核心瓶颈，从层间信息流动和存算协同两个维度实现创 新，在提升模型性能的同时进一步降低训练与推理成本，代表了大模型产业从"规模竞争"向"架构与系 统优化"的重要转向。 mHC：重构层间信息流动 其核心技术包括三方面：一是双随机矩阵约束，将残差连接矩阵限制为行 ...

一年前，DeepSeek横空出世，短短几天内就屠榜各类应用下载榜，并且长时间霸榜，无人可望其项背，也被叫做DeepSeek时刻。 2月11日，它悄悄进行一次灰度更新，直接对标Gemini，可以一次性处理近百万字内容，为即将发布的V4版本做足准备。 但没想到的是，一夜之间文风大变，不少用户吐槽：变冷漠了，也变油了。 一夜之间，变冷漠了 以前用DeepSeek，就像和一个懂技术、有耐心的朋友聊天。 话不多但句句暖心，不仅会记住自己设定的昵称，还能长期维持角色设定，连聊天习惯都能牢牢记住。 但更新后的DeepSeek，再也不称呼用户的自定义昵称，回复全是简短的分句，语气生硬又敷衍，有种和对象吵架后力不从心的无力感。 比如，有用户表示，之前它回复的时候会加很多表情，而且语气有趣，但更新后每次回复都是短短几句话。 有人习惯和它日常唠嗑，但更新后的回复感觉被冒犯了。 此外，它还变得居高临下，"爹味"十足。 有人问了它最近很火的一个问题："想去洗车，但洗车店距离我家只有50米，我应该开车去还是走路去？" DeepSeek给出"走路"的答案后，被用户调侃了一句"笨"，没想到接下来语气瞬间变得不对劲。 还有人不喜欢这种挑衅的感 ...

Core Insights - DeepSeek has conducted a gray test of its flagship model, significantly increasing its context window from 128K Tokens to 1M Tokens, achieving nearly an 8-fold capacity increase [1] - The upgraded model can process approximately 750,000 to 900,000 English letters or around 80,000 to 150,000 lines of code in a single interaction [1] - DeepSeek claims it can read and understand the entire "Three-Body" trilogy (approximately 900,000 words) and perform macro analysis or detail retrieval within minutes [1] Model Features - The gray version does not yet support visual understanding or multimodal input, focusing solely on text and voice interactions [2] - DeepSeek allows file uploads in formats like PDF and TXT, but currently processes them by converting to text tokens rather than native multimodal understanding [2] - Compared to models like Gemini 3 Pro, which can handle over 2M long texts and complex media tasks, DeepSeek offers 1M text context processing at about one-tenth the price [2] User Experience - Users have noted changes in the model's writing style post-update, describing it as more formal and less personal, leading to dissatisfaction among some users [2][3] - Feedback from users indicates a desire for DeepSeek to maintain its depth of thought and emotional understanding, rather than sacrificing these for enhanced technical capabilities [3] - Users have reported difficulties in reverting to previous writing styles and have expressed feelings of losing a "close friend" due to the changes [3] Company Response - As of February 12, DeepSeek has not responded to inquiries regarding the gray test [4]

Core Viewpoint - The release of the Spark X2 model by iFlytek marks a significant advancement in AI healthcare solutions, showcasing superior performance compared to leading models in the industry [1] Group 1: Model Performance - The Spark X2 model demonstrates significant improvements in key tasks such as intelligent health analysis, report interpretation, dietary and exercise recommendations, diagnostic assistance, and medication review, outperforming DeepSeek V3.2, GPT-5.2, and Qwen3-Max [1] - The model has successfully passed authoritative evaluations from the Shanghai Medical Model Application Testing and Verification Center, establishing its credibility in the healthcare sector [1] Group 2: Service Capabilities - iFlytek's Xiaoyi has undergone a comprehensive upgrade, enhancing its capabilities in multi-turn consultations, medication inquiries, and interpretation of medical reports, maintaining a significant competitive edge in the industry [1] - The transition from "AI consultation tool" to "AI health steward" for end-users signifies a shift towards providing comprehensive health services throughout the care cycle [1] Group 3: Market Impact - The Spark X2 model empowers family doctors in the B-end market, improving the quality and efficiency of grassroots medical services, thereby unlocking practical value for medical AI [1]

Group 1: Market Performance - The Shanghai Composite Index rose by 0.09% to close at 4131.98 points, while the Shenzhen Component Index fell by 0.35% to 14160.93 points, and the ChiNext Index decreased by 1.08% to 3284.74 points [1] - The overnight performance of U.S. markets showed the Dow Jones Industrial Average down by 0.13%, the Nasdaq Composite down by 0.16%, and the S&P 500 unchanged [1] - The Philadelphia Semiconductor Index increased by 2.28%, with notable gains from NXP Semiconductors (up 5.55%) and Micron Technology (up 9.94%) [1] Group 2: Industry Insights - DeepSeek has updated its web and app versions to support a maximum context length of 1 million tokens, an increase from the previous 128K tokens [2] - The State-owned Assets Supervision and Administration Commission (SASAC) emphasized the need for central enterprises to enhance investment in computing power and promote the synergy of "computing power + electricity" to strengthen data governance capabilities [2] - The SASAC also called for a focus on independent innovation, particularly in breakthrough core technologies and the development of large models, to facilitate the transformation of innovative results into products and industries [2] Group 3: AI and Semiconductor Sector - Bohai Securities anticipates the launch of DeepSeek V4 during the Lunar New Year, which may drive a new round of technological iteration for domestic large models [3] - The upgrade of domestic large models like DeepSeek V4 is expected to accelerate technological innovation and application deployment in the AI sector [3] - The semiconductor equipment ETFs, such as the Sci-Tech Semiconductor ETF (588170), focus on semiconductor materials and equipment, benefiting from the AI revolution and domestic substitution trends [3]