[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fip71gZpI77oAOgWwcWJoYc7mrwgxpm7fsIDuBJiSAXU":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},2032,"我国团队研发新一代大原子模型 DPA4 性能登顶国际权威榜单","\u003Cp>\u003Cspan style=\"font-size: 18px;\">2026 年 5 月，北京科学智能研究院联合北京大学、北京应用物理与计算数学研究所团队，发布新一代大原子模型\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\">DPA4\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">，在国际权威榜单实现双榜登顶，综合性能位居全球第一。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">DPA4 在\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\">Matbench Discovery\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">（无机材料 AI 发现领域国际金标准）上综合性能指标世界第一；同时在分子领域权威基准\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\">SPICE-MACE-OFF\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">榜单夺冠，成为通用势能面模型领域当前最优（SOTA）模型。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">该模型最大亮点是\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\">高精度、低成本、高效率\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">。相比此前领先模型，DPA4 参数量不足其十分之一，训练成本大幅降低，单张消费级显卡仅需约一天即可完成训练；精度不变前提下，训练效率较上一代 DPA3 提升约 10 倍。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">技术核心在于 “化繁为简”：团队为原子间 “边” 建立局部坐标系，把复杂全局旋转简化为固定轴旋转，再引入注意力机制，用更少参数承载更丰富信息，实现性能与成本的双重突破。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">DPA4 应用场景广泛，覆盖催化剂、电池材料、半导体材料、药物分子等多领域研究，还可作为干湿闭环自主实验室的基座模型。目前 DPA4 已开放社区尝鲜版，后续将开源论文与正式版本，推动领域开放共建。此次突破，标志我国在 AI 材料计算领域跻身全球领先行列。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>【新闻来源】人民网   http:\u002F\u002Ffinance.people.com.cn\u002Fn1\u002F2026\u002F0526\u002Fc1004-40727256.html\u003C\u002Fp>\u003Cp>（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F05\u002F7d5ee26db82048ec9ceaf1e5c6ae6be2\u002F1780022346308.png","https:\u002F\u002Fimage.51xinwei.com\u002F2026\u002F05\u002Fthumbs\u002F7d5ee26db82048ec9ceaf1e5c6ae6be2\u002F1780022346308.png",0,1,28,"2026-05-29 10:39",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3Aebe6d4b0-33c9-4edd-9b35-d2c7a76567b1%3A0.wav?Expires=1780026029&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=F1sUnfbdO94rI%2BzHivWt0bG3xU8%3D",2963640,"ebe6d4b0-33c9-4edd-9b35-d2c7a76567b1","2026-05-29 10:34","Chinese Team Develops New-Generation Large Atomic Model DPA4, Topping Performance on International Authoritative Benchmark","\u003Cp>\u003Cspan style=\"font-size: 18px;\">In May 2026, the Beijing Institute for Scientific Intelligence, in collaboration with teams from Peking University and the Beijing Institute of Applied Physics and Computational Mathematics, released the new-generation large atomic model\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\"> DPA4\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">, achieving top positions on two authoritative international benchmarks and securing the number one spot globally for overall performance.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">DPA4 achieved world-leading comprehensive performance metrics on\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\"> Matbench Discovery\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\"> (the international gold standard in AI discovery for inorganic materials); simultaneously, it claimed the top spot on the authoritative molecular benchmark\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\"> SPICE-MACE-OFF\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">, making it the current state-of-the-art (SOTA) model in the field of universal potential energy surface models.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">The model's most notable highlights are its\u003C\u002Fspan>\u003Cstrong style=\"background-color: rgba(0, 0, 0, 0); font-size: 18px;\"> high accuracy, low cost, and high efficiency\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\">. Compared to the previous leading model, DPA4 has less than one-tenth of the parameters, significantly reducing training costs; training can be completed in approximately one day using a single consumer-grade GPU. While maintaining the same level of accuracy, its training efficiency is about 10 times higher than the previous generation DPA3.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">The technical core lies in \"simplifying complexity\": the team established a local coordinate system for the \"edges\" between atoms, simplifying complex global rotations into fixed-axis rotations, and then introduced an attention mechanism. This allows richer information to be carried with fewer parameters, achieving breakthroughs in both performance and cost.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">DPA4 has broad application scenarios, covering research in multiple fields such as catalysts, battery materials, semiconductor materials, drug molecules, and can also serve as a foundational model for autonomous wet\u002Fdry closed-loop laboratories. Currently, a community preview version of DPA4 has been released, with the research paper and the official version to be open-sourced subsequently to promote open collaboration in the field. This breakthrough marks China's entry into the global leading ranks in the field of AI materials computing.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>[News Source] People's Daily Online   http:\u002F\u002Ffinance.people.com.cn\u002Fn1\u002F2026\u002F0526\u002Fc1004-40727256.html\u003C\u002Fp>\u003Cp>(This website republishes this article to provide readers with more information. The content does not constitute investment or consumption advice. If there are questions regarding the facts in the article, please verify with the relevant parties. The views expressed in the article are not necessarily those of this website and are for readers' reference only.)\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A656cfd65-b2dd-469d-affa-b22355b85ca6%3A0.wav?Expires=1780039188&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=x3EwXWXiWYJSeJV1m3nHbJlWgSc%3D","656cfd65-b2dd-469d-affa-b22355b85ca6",4784510]