据悉，在维持极高数据压缩效率的同时，DeepSeek-OCR 2 在多项基准测试和生产指标上均取得了显著 突破。模型仅需 256 到 1120 个视觉 Token 即可覆盖复杂的文档页面，这在同类模型中处于极低水平， 显著降低了下游 LLM 的计算开销。在 OmniDocBench v1.5 评测中，其综合得分达到 91.09%，较前代提 升了 3.73%，特别是在阅读顺序识别方面表现出了更强的逻辑性。 责任编辑：宋雅芳 责任编辑：宋雅芳 新浪科技讯 1月27日下午消息，DeepSeek团队今日《DeepSeek-OCR 2： Visual Causal Flow》论文并开 源了DeepSeek-OCR 2模型。据悉，该模型采用创新的DeepEncoder V2架构，实现了视觉编码从固定扫 描向语义推理的范式转变，可让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。 新浪科技讯 1月27日下午消息，DeepSeek团队今日《DeepSeek-OCR 2： Visual Causal Flow》论文并开 源了DeepSeek-OCR 2模型。据悉，该模型采用创新的DeepEncoder ...

Group 1 - DeepSeek's new DeepSeek-OCR 2 model utilizes the innovative DeepEncoder V2 method, enabling AI to "see" an image in a logical sequence similar to humans [2] - The AI ETF has seen a net inflow of 1.375 billion yuan over the past 10 days, with a current scale of 10.961 billion yuan, covering various segments of the AI industry chain [3] - The robot-themed ETF includes leading companies such as Huichuan Technology, leading in industrial robots, and Stone Technology, a leader in service robots [3] Group 2 - The Yuangbao plan will distribute 1 billion yuan in cash red envelopes starting February 1, with individual envelope amounts reaching up to 10,000 yuan [2] - Wenxin will launch a cash red envelope campaign totaling 500 million yuan from January 26 to March 12, also with a maximum prize of 10,000 yuan [2] - ByteDance's Volcano has become the exclusive AI cloud partner for CCTV's Spring Festival Gala, indicating potential for further promotional activities during the event [2]

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]

每经AI快讯，1月27日消息，DeepSeek发布全新DeepSeek-OCR2模型，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，而不再只是机械地从左到右扫描。这种方式 模拟了人类在观看场景时所遵循的逻辑流程。最终，该模型在处理布局复杂的图片时，表现优于传统的 视觉-语言模型，实现了更智能、更具因果推理能力的视觉理解。 ...

风险提示及免责条款 市场有风险，投资需谨慎。本文不构成个人投资建议，也未考虑到个别用户特殊的投资目标、财务状况或需要。用户应考虑本文中的任何 意见、观点或结论是否符合其特定状况。据此投资，责任自负。 DeepSeek发布全新DeepSeek-OCR 2模型，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，而不再只是机械地从左到右扫描。这种方式模拟了人类在观看场景时所遵循 的逻辑流程。最终，该模型在处理布局复杂的图片（如文档或图表）时，表现优于传统的视觉-语言模 型，实现了更智能、更具因果推理能力的视觉理解。 ...

一年前，恰是这个周一，深度求索（DeepSeek）公司R1大模型以高性能和低训练成本震撼世界，外媒 称之为"DeepSeek时刻"。 DeepSeek的横空出世，可谓2025年中国科技创新的"开场戏"。近期，国际机构密集发文盘点过去一年全 球科技发展，来自中国的成果格外耀眼。 在基础研究领域，2025年底，国际顶尖学术期刊《科学》公布"2025年度十大科学突破"，中国在可再生 能源领域的突破，以及破解"丹尼索瓦人"长相之谜、发现水稻耐高温"基因开关"等成果纷纷入选。 在人工智能（AI）领域，据麻省理工学院等机构2025年11月底统计，DeepSeek、千问等中国开源模型 下载量已占全球总量的17%，超越美国位居全球第一。今年1月，微软发布报告称，DeepSeek在俄罗 斯、白俄罗斯、伊朗、非洲等发展中国家的普及率呈"爆炸式"增长。 在与AI算力息息相关的先进芯片领域，全球知名金融研究机构伯恩斯坦研究公司预测，2026年华为等 本土企业将占据中国AI芯片市场80%的份额，英伟达的市场份额将降至8%。 过去一年，中国科技创新因何能让世界瞩目？ 这源于实现高水平科技自立自强的决心与努力。在美国先进芯片的封锁禁运下， ...

（来源：吴说） 吴说获悉，据 Top Ledger 数据，自 Solana Mobile 空投以来，Seeker 代币 SKR 累计交易量已超过 2 亿美 元；现约 85% 的空投总额已被领取；SKR 的链上交易量主要在 Meteora 中完成，占据了 57.4%。 ...

生意社01月26日讯 1月26日，山东万华化学，华东地区LLDPE7042报7050元/吨，上涨200元/吨；华东地区LLDPE7050报 7100元/吨，上涨200元/吨。 PriceSeek评析 LLDPE，多空评分：2 生意社01月26日讯 1月26日，山东万华化学，华东地区LLDPE7042报7050元/吨，上涨200元/吨；华东地区LLDPE7050报 7100元/吨，上涨200元/吨。 PriceSeek评析 LLDPE，多空评分：2 文章显示华东地区LLDPE现货报价上涨200元/吨，如7042型号报7050元/吨、7050型号报7100元/吨，涨 幅显著。这反映出供应紧张或需求增加，对现货价格构成重大利好。结合大连商品交易所聚乙烯期货数 据，2605合约收盘价6865元/吨，上涨116元/吨，成交量达719797手，持仓量增加3532手，表明市场多 头情绪浓厚。现货价格上涨将强化期货上行预期，支撑未来价格走势。 【大宗商品公式定价原理】生意社基准价是基于价格大数据与生意社价格模型产生的交易指导价，又称 生意社价格。可用于确定以下两种需求的交易结算价： 1、指定日期的结算价 2、指定周期的平均结 ...

但如果把问题继续往深处追问：推理能力的本质，真的只是多算几步吗？ 谷歌、芝加哥大学等机构的研究者最近发表的一篇论文给出了一个更具结构性的答案，推理能力的提升并非仅源于计算步数的增加，而是来自模型在推理 过程中隐式模拟了一种复杂的、类多智能体的交互结构，他们称之为「思维社会」（society of thought）。 过去两年，大模型的推理能力出现了一次明显的跃迁。在数学、逻辑、多步规划等复杂任务上，推理模型如 OpenAI 的 o 系列、DeepSeek-R1、QwQ- 32B，开始稳定拉开与传统指令微调模型的差距。直观来看，它们似乎只是思考得更久了：更长的 Chain-of-Thought、更高的 test-time compute，成为最常 被引用的解释。 简单理解就是，这项研究发现，为了解决难题，推理模型有时会模拟不同角色之间的内部对话，就像他们数字大脑中的辩论队一样。他们争论、纠正对 方、表达惊讶，并调和不同观点以达成正确答案。人类智能很可能是因为社交互动而进化的，而类似的直觉似乎也适用于人工智能！ 通过对推理输出进行分类，以及结合作用于推理轨迹的机制可解释性方法，研究发现，诸如 DeepSeek-R ...

Core Viewpoint - The discussion centers around the competitive landscape in the AI field, particularly focusing on the contrasting approaches of Mistral and DeepSeek in developing sparse mixture of experts (MoE) models, with Mistral's CEO acknowledging China's strong position in AI and the significance of open-source models [1][4]. Group 1: Company Perspectives - Mistral's CEO, Arthur Mensch, claims that open-source models are a strategy for progress rather than competition, highlighting their early release of open-source models [1]. - The recent release of DeepSeek-V3 is built on Mistral's proposed architecture, indicating a collaborative yet competitive environment in AI development [1][4]. - There is skepticism among the audience regarding Mistral's claims, with some suggesting that Mistral's recent models may have borrowed heavily from DeepSeek's architecture [4][13]. Group 2: Technical Comparisons - Both DeepSeek and Mistral's Mixtral focus on sparse MoE systems, aiming to reduce computational costs while enhancing model capabilities, but they differ fundamentally in their approaches [9]. - Mixtral emphasizes engineering principles, showcasing the effectiveness of a robust base model combined with mature MoE technology, while DeepSeek focuses on algorithmic innovation to address issues in traditional MoE systems [9][12]. - DeepSeek introduces a fine-grained expert segmentation approach, allowing for more flexible combinations of experts, which contrasts with Mixtral's flat knowledge distribution among experts [11][12]. Group 3: Community Reactions - The community has reacted critically to Mistral's statements, with some users expressing disbelief and pointing out the similarities between Mistral's and DeepSeek's architectures [2][17]. - There is a sentiment that Mistral, once a pioneer in the open-source AI space, is now perceived as having lost its innovative edge, with DeepSeek gaining more influence in the sparse MoE and MLA technologies [14][17]. - The competitive race for foundational models is expected to continue, with DeepSeek reportedly targeting significant releases in the near future [19].