Group 1 - The core viewpoint of the article highlights that by March 2025, the monthly active users of mobile AI applications in China will exceed 647 million, indicating that over half of Chinese internet users have entered the "AI-native application" era, with educational AI emerging strongly in this technological wave [1] - The success of Xiaoyuan AI, designed specifically for primary and secondary school students, is attributed to a deep understanding of the essence of education, as evidenced by its top position in user growth within the education AI sector and its entry into the top 10 newly downloaded AI applications across the internet [1][2] - Xiaoyuan AI is built on a robust dataset accumulated by Yuanfudao Group over more than a decade, which includes over 5 million hours of teaching videos, more than 2 billion question bank entries, over 10,000 knowledge points, and 1 million exam papers, along with 30 billion learning data points from 500 million global users, providing a solid foundation for high-precision learning analysis and personalized learning path planning [2][3] Group 2 - The core layer of Xiaoyuan AI integrates "educational genes" with "technological evolution" through the collaboration of self-developed Yuanli model and the top reasoning model Deepseek-R1, creating a unique model matrix that accurately identifies the root causes of student errors and transforms hard data into guided teaching that aligns with cognitive principles [3][5] - Xiaoyuan AI APP offers a comprehensive learning solution for primary and secondary students, covering over 100 key learning scenarios such as homework checking, learning diagnostics, error analysis, and 1v1 explanations, with homework correction serving as a core scenario and the 1v1 personalized explanation feature achieving significant penetration in the past month, becoming a trusted learning assistant for parents and students [5]

Core Viewpoint - Tencent has integrated its AI applications with the DeepSeek R1-0528 model, allowing users to experience advanced capabilities in deep thinking, programming, and long text processing across various platforms for free and without limits [1] Group 1: AI Application Integration - Multiple Tencent AI applications, including Tencent Yuanbao, ima, Sogou Input Method, QQ Browser, Tencent Docs, Tencent Maps, and Tencent LeXiang, have announced the integration with DeepSeek R1-0528 [1] - Users can select the DeepSeek model R1 for enhanced functionalities across different products [1] Group 2: Cloud Services - Tencent Cloud has launched DeepSeek-R1-0528, enabling enterprises and developers to access the API interface for stable and high-quality services [1] - The Tencent Cloud Intelligent Agent Development Platform offers built-in capabilities for RAG, workflow, and agent development, facilitating the rapid creation of customized intelligent applications [1] - Tencent Cloud's TI platform allows for fine-tuning of the model, enhancing its adaptability for specific use cases [1]

Core Insights - DeepSeek's R1 model has undergone a minor version upgrade, enhancing semantic understanding, complex logical reasoning, and long text processing stability [1] - The upgraded model shows significant improvements in understanding capabilities and programming skills, capable of generating over 1000 lines of error-free code [1] - The R1 model's cost-effectiveness is highlighted, being priced at 1/11 of Claude-3.7-Sonnet and 1/277 of GPT-4.5, while being open-source for commercial use [1] Group 1 - The R1 model has gained global attention since its January release, outperforming Western competitors and causing a drop in tech stocks [2] - Following the release of the V3 model, interest in DeepSeek has shifted towards the anticipated R2 model, which is expected to utilize a mixture of experts model with 1.2 trillion parameters [2] - The latest version R1-0528 has sparked renewed media interest, showcasing competitive performance against OpenAI's models in code generation [2] Group 2 - DeepSeek's low-cost, high-performance R1 model has positively influenced the Chinese tech stock market and reflects optimistic market expectations regarding China's AI capabilities [2] - The upgrade has also shown improvements in reducing hallucinations, indicating that DeepSeek is not only catching up but competing with top models [1]

Group 1 - The Trump administration's tariffs faced legal challenges, with a federal appeals court temporarily halting a lower court's ruling that blocked several tariff orders [2] - The U.S. stock market showed positive performance, with the Dow Jones up 0.28%, Nasdaq up 0.39%, and S&P 500 up 0.4% [3][5] - Major Chinese concept stocks mostly rose, with the Nasdaq China Golden Dragon Index increasing by 1.44% [4] Group 2 - The White House indicated that a judge's ruling on tariffs would be overturned, and multiple trade agreements are nearing completion [8] - The Federal Reserve's Goolsbee suggested that if trade policies revert to pre-tariff conditions, there could be room for interest rate cuts [9][10] - Goldman Sachs predicted gold prices could reach $4,000 per ounce by mid-next year, viewing gold as a safer hedge compared to Bitcoin [11] Group 3 - Nvidia's CEO plans to sell up to 6 million shares, potentially worth around $809 million based on recent closing prices [12] - Dell Technologies saw a significant stock increase after reporting Q1 revenue of $23.38 billion, exceeding analyst expectations [13] - Hyundai is considering a 1% price increase on all its U.S. products to mitigate the impact of Trump’s tariffs [14] Group 4 - NIO reported Q1 revenue of 25.93 billion yuan, a year-on-year increase of 1.1%, with a delivery volume of 92,864 vehicles, up 15.5% [17] - JD.com and Xiaohongshu launched a strategic cooperation plan to enhance business growth for brands and merchants [18] - Tesla plans to deliver its first autonomous Model Y vehicles in June, ahead of schedule [19]

每经记者｜可杨 每经编辑｜张海妮 丨 2025年5月30日 星期五 丨 NO.1 宇树科技从有限公司变更为股份公司 5月29日，宇树科技向合作伙伴发布通知称，因公司发展需要，杭州宇树科技有限公司即日起名称变更 为"杭州宇树科技股份有限公司"。原公司所有业务由"新公司"继续经营，原公司签订的所有合同继续有 效。 点评：宇树科技从有限责任公司到股份有限公司的转身，是企业自身发展壮大的必然选择，也是科技行 业创新发展的生动缩影。在新的股份制架构下，期待宇树科技凭借更强大的资本实力、更灵活的运营机 制与更高效的治理结构，在科技领域开启新的征程。 大模型明星企业DeepSeek深夜"上新"。5月29日凌晨，DeepSeek开源了R1最新0528版本。DeepSeek目前 没有对该版本进行任何说明，只是"悄悄"地开放了模型。著名代码测试平台Live CodeBench显示，其性 能可以媲美OpenAI最新的o3模型的高版本。也有网友对新版R1的风格进行了测试，结果几乎和OpenAI 的o3差不多。 点评：开源是推动技术进步和生态发展的重要方式。DeepSeek R1新版本的开源，为开发者提供了更多 选择和创新机会，有助于 ...

Summary of Deepseeker R1 Conference Call Company and Industry - The discussion revolves around the performance and updates of the Deepseeker R1 model, a product in the AI and machine learning industry, particularly focusing on its capabilities in data retrieval and code generation. Core Points and Arguments - **Performance Improvement**: The accuracy of Deepseeker R1 in CLion improved from 4/8 to 6/8 in version 0.528, although it still lags behind Claude 3.7 (7/8) and CosmoFlow with Claude 4 (8/8) [1][3][19]. - **Context Length Enhancement**: The new version increased the maximum context length to 128K for clients, addressing previous issues where excessive web content retrieval exceeded context limits [5][19]. - **Challenges in Data Retrieval**: The model faced difficulties using the fetch tool to retrieve China’s GDP data due to low success rates and lack of API support from the World Bank, indicating compatibility issues between MCP tools and large models [6][19]. - **Comparison with Other Models**: Readcloud 3.7, Readcloud 4, Grok 3, and Gemini 2.5 Pro demonstrated better performance in using MCP tools and parameter settings, successfully completing tasks that Deepseeker R1 struggled with [7][19]. - **Code Generation Quality**: While the new version shows improvements in reasoning and text generation quality, the code generation aspect still has flaws compared to Claude series models [4][19]. - **Error Handling in MCP Tools**: The MCP tools often encounter issues when a tool fails, and the selection of alternatives is not always ideal. Readcloud has shown the ability to quickly find substitutes when issues arise [13][14]. Other Important but Possibly Overlooked Content - **Task Complexity**: The complexity of tasks requiring multiple MCP tools can lead to cascading errors if one tool fails, emphasizing the need for careful planning and tool selection [11][19]. - **Improvements in Cloud 4**: Cloud 4 outperforms Cloud 3.7 in data scraping and webpage generation, with faster speeds and higher accuracy, showcasing advancements in the technology [10][19]. - **Devsec Error Handling**: Devsec's error handling is contingent on initial tool selection, suggesting a need for improved recognition and selection of backup options to enhance reliability [15][19]. - **Limitations in Code Generation**: Despite improvements, the new version's code generation still falls short in quality compared to Claude 3.7 and 4, particularly in achieving expected outcomes in specific projects [17][19]. - **Overall Model Comparison**: Claude 4 is noted for its superior speed and accuracy, especially in programming tasks, indicating a competitive edge over Deepseeker R1 [18][19].

Core Insights - The updated R1 model from DeepSeek has significantly improved its capabilities, particularly in reducing the "hallucination" rate, which previously stood at around 21% [1][4]. Model Performance - The new R1 model has achieved top-tier performance in various benchmark tests, surpassing all domestic models and nearing the performance of international leaders like o3 and Gemini-2.5-Pro [4]. - The hallucination rate has been reduced by approximately 45%-50% in tasks such as rewriting, summarization, and reading comprehension, providing more accurate and reliable results [4][18]. - In the AIME 2025 test, the model's accuracy improved from 70% to 87.5% in complex reasoning tasks [18]. Model Features and Capabilities - The updated R1 model can generate longer and more structured pieces of writing, including essays, novels, and prose, while aligning more closely with human writing styles [18]. - The model's coding capabilities have also seen significant enhancements, performing nearly on par with OpenAI's o3-high model in code testing environments [18]. - The new model has a parameter count of 685 billion and supports a context length of 128K in the open-source version [19]. Future Developments - There is considerable anticipation in the industry for the next-generation R2 model, with users expressing their eagerness for its release [19]. - DeepSeek has not commented on speculations regarding the R2 model, but the ongoing competition in the foundational model space remains intense [19].

5月29日晚间，深度求索微信公众号公布了 DeepSeek-R1-0528 更新的详细升级内容，DeepSeek-R1-0528 仍然使用 2024年12月所发布的 DeepSeek V3 Base 模型作为基座，但在后训练过程中投入了更多算力，显著提升了模型的思维深度与推理能力。更新后的 R1 模型在数学、编程与通用逻辑 等多个基准测评中取得了当前国内所有模型中首屈一指的优异成绩，并且在整体表现上已接近其他国际顶尖模型，如 o3 与 Gemini-2.5-Pro。 其他能力更新比如： 1.幻觉改善：新版 DeepSeek R1 针对"幻觉"问题进行了优化。与旧版相比，更新后的模型在改写润色、总结摘要、阅读理解等场景中，幻觉率降 低了 45～50% 左右，能够有效地提供更为准确、可靠的结果。 2.创意写作：在旧版 R1 的基础上，更新后的 R1 模型针对议论文、小说、散文等文体进行了进一步优化，能够输出篇幅更长、结构内容更完整的 长篇作品，同时呈现出更加贴近人类偏好的写作风格。 3.工具调用：DeepSeek-R1-0528 支持工具调用（不支持在 thinking 中进行工具调用）。当前模型 Tau-Ben ...

这一进步得益于模型在推理过程中的思维深度增强：在 AIME 2025 测试集上，旧版模型平均每题使用 12K tokens，而新版模型平均每题使用 23K tokens， 表明其在解题过程中进行了更为详尽和深入的思考。 此外，新版 DeepSeek R1 针对"幻觉"问题进行了优化。与旧版相比，更新后的模型在改写润色、总结摘要、阅读理解等场景中，幻觉率降低了 45～50% 左右，能够有效地提供更为准确、可靠的结果。在旧版 R1 的基础上，更新后的 R1 模型针对议论文、小说、散文等文体进行了进一步优化，能够输出篇 幅更长、结构内容更完整的长篇作品，同时呈现出更加贴近人类偏好的写作风格。 | Benchmarks | DeepSeek-R1- | OpenAI- | Gemini-2.5- | Qwen3- | DeepSeek-R1 | | --- | --- | --- | --- | --- | --- | | | 0528 | 03 | Pro-0506 | 235B | | | AIME 2024 数学竞赛 pass@1 | 91.4 | 91.6 | 90.8 | 85.7 | 79.8 | | A ...

新浪科技讯 5月29日晚间消息，DeepSeek今日宣布，DeepSeek R1模型已完成小版本升级，当前版本为 DeepSeek-R1-0528。用户通过官方网站、App或小程序进入对话界面后，开启"深度思考"功能即可体验 最新版本。API 也已同步更新，调用方式不变。 工具调用，DeepSeek-R1-0528 支持工具调用（不支持在 thinking 中进行工具调用）； 据介绍，DeepSeek-R1-0528 仍然使用 2024 年 12 月所发布的 DeepSeek V3 Base 模型作为基座，但在后 训练过程中投入了更多算力，显著提升了模型的思维深度与推理能力。官方称更新后的 R1 模型在数 学、编程与通用逻辑等多个基准测评中取得了当前国内所有模型中首屈一指的优异成绩，并且在整体表 现上已接近其他国际顶尖模型，如o3与Gemini-2.5-Pro。 其他能力更新方面，包括幻觉改善，新版 DeepSeek R1 针对"幻觉"问题进行了优化。与旧版相比，更新 后的模型在改写润色、总结摘要、阅读理解等场景中，幻觉率降低了45~50%左右，能够有效地提供更 为准确、可靠的结果； 创意写作，在旧版 R1 ...