[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fF3N0QTpo0TX7j5hXFdCWC7k4R6dk2cPga4z8nRb0MJA":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},1202,"AI领域迎来重大突破！GPT-5震撼发布，性能全面升级、多领域“屠榜”，告别“幻觉”？","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px;\">北京时间8月8日凌晨，OpenAI正式发布了备受瞩目的新一代人工智能模型GPT-5。这场长达1个多小时的线上发布会规格之高、细节之多，引发了全球科技界的广泛关注。GPT-5的发布，不仅标志着AI技术的一次巨大飞跃，更被视为构建“超级智能”的重要一步。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">性能全面升级，多领域“屠榜”\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5在多个领域的表现都迎来了全面提升。在文本生成、Web开发和视觉感知能力领域，GPT-5均排名第一；在硬提示、编码、数学、创造力、长查询等方面也毫无悬念地拔得头筹。在代号为“峰会”的测试中，GPT-5保持着截至目前最高的Arena分数，堪称实至名归的“屠榜”者。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">OpenAI首席执行官Sam Altman对GPT-5的评价极高，他比喻说，如果说GPT-4像中学生，GPT-5就像大学生，而GPT-5则是第一次让他真正感觉像在与一位博士级别的专家交谈。他甚至断言，GPT-5是世界上编码能力最强、写作能力最强以及在医疗保健领域最强的模型。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">告别“幻觉”，AI更可靠\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“AI幻觉”一直是困扰大型语言模型的难题，但GPT-5在这方面取得了显著进步。与GPT-4o相比，GPT-5在联网搜索时回答出现事实错误的概率降低了45%；在独立思考时，回答出错的概率更是比OpenAI o3降低了80%。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">此外，GPT-5还引入了四种全新的“人格”模式，包括犬儒、机器人、倾听者和学霸，用户可以根据自己的喜好设定ChatGPT与自己互动和回答问题的方式。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">编码能力逆天，“软件按需生成”时代来临\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5的编码能力堪称逆天。在多个编码基准测试中，GPT-5的表现均优于其他任何模型。在人类最终测试中取得42%的成绩，在SWE基准测试中更是达到了75%的成绩。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在发布会上，OpenAI的后期训练负责人Yann Dubois现场演示了GPT-5的强大编码能力，要求它生成一个用于学习法语并带有互动游戏的网站。GPT-5在短短几秒钟内就编写了数百行代码，并直接展示了网站的前端界面。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">此外，GPT-5还展示了一款仅靠一段提示词就创作的3D游戏，其画面精致，物理效果也十分逼真。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">更安全、更“诚实”\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">OpenAI对GPT-5进行了超过五千小时的测试，以确保其安全性。GPT-5在更可靠地处理多步骤任务方面表现更好，不会像以前那样“自信地撒谎”。对于一些可能涉及安全风险的问题，GPT-5会采用“安全补全”机制，在保持安全约束的前提下，给出尽可能有帮助的答案。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">全面开放，多版本满足不同需求\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5向所有用户开放，包括普通用户、Plus会员和Pro会员。普通用户可以免费体验GPT-5，但用量有限；Plus会员可获得更多使用量；Pro会员则可以访问具备更强推理能力的GPT-5 Pro版本。GPT-5系列还包括GPT-5-mini、GPT-5-nano等版本，以满足不同用户的需求。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在定价方面，标准版GPT-5每百万输入Token 1.25美元，每百万输出Token 10美元，mini版和Nano版的定价则相对较低。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">集成架构，智能调度子模型\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5采用了内嵌式三位一体集成架构，由处理常规问题的GPT-5-main模型、解决复杂任务的GPT-5-thinking深度思考模型、实时决策的路由机制以及额度使用完后启动的mini版本组成。路由机制会根据对话类型、复杂程度、工具需求以及用户的明确意图，快速决定使用哪个模型，自动匹配最适合的模型，用户无需手动切换。\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>\u003Cspan style=\"color: rgb(187, 187, 187);\">【新闻来源】金融界财经 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002F%E6%8A%80%E6%9C%AF\u002F%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD\u002Fai%E9%A2%86%E5%9F%9F%E8%BF%8E%E6%9D%A5%E9%87%8D%E5%A4%A7%E7%AA%81%E7%A0%B4-gpt-5%E9%9C%87%E6%92%BC%E5%8F%91%E5%B8%83-%E6%80%A7%E8%83%BD%E5%85%A8%E9%9D%A2%E5%8D%87%E7%BA%A7-%E5%A4%9A%E9%A2%86%E5%9F%9F-%E5%B1%A0%E6%A6%9C-%E5%91%8A%E5%88%AB-%E5%B9%BB%E8%A7%89\u002Far-AA1K7ABJ?ocid=BingNewsSerp\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">http:\u002F\u002Fu5a.cn\u002FgVkVC\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（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\u002Fspan>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002F12baa7de2eaa41a8a58c094c73e6375b\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F08\u002Fthumbs\u002F12baa7de2eaa41a8a58c094c73e6375b\u002FAI领域.jpg",0,1,227,"2025-08-08 17:48",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%3Ab0988099-92a9-4670-9d6f-04841ccb7131%3A0.wav?Expires=1754652782&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=ZzUy4tiUGR0o2KXApMG07F8EKUQ%3D",7491752,"b0988099-92a9-4670-9d6f-04841ccb7131","2025-08-08 17:43","Major breakthrough in the AI field! GPT-5 is officially launched, with performance fully upgraded and dominating multiple fields. Is it free from \"hallucinations\"?","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px;\">At 08:00 Beijing Time on August 8th, OpenAI officially released its highly anticipated new artificial intelligence model, GPT-5. This online launch event lasting more than an hour was of high specification and rich in details, drawing widespread attention from the global technology community. The release of GPT-5 not only marks a major leap in AI technology, but is also seen as an important step in building \"super intelligence\".\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">Performance Fully Upgraded, Dominating Multiple Fields\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5 has achieved comprehensive improvements in performance across multiple fields. It ranks first in text generation, web development, and visual perception capabilities; it also leads without any doubt in hard prompts, coding, mathematics, creativity, and long queries. In the test named \"Summit\", GPT-5 maintains the highest Arena score so far, making it a true \"dominator\" of rankings.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">OpenAI CEO Sam Altman has a very high opinion of GPT-5. He compared it by saying that if GPT-4 is like a high school student, GPT-5 is like a college student, and GPT-5 is the first time he truly feels like talking to an expert at the doctoral level. He even claimed that GPT-5 is the strongest model in coding, writing, and healthcare fields in the world.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">Say Goodbye to \"Hallucinations\", AI Becomes More Reliable\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\"AI hallucinations\" have always been a problem for large language models, but GPT-5 has made significant progress in this area. Compared to GPT-4o, the probability of factual errors in GPT-5's answers when searching online has decreased by 45%; when thinking independently, the error rate is even 80% lower than that of OpenAI o3.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In addition, GPT-5 introduced four new \"personality\" modes, including cynical, robot, listener, and学霸 (academic star), users can set the way ChatGPT interacts and answers questions according to their preferences.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">Coding Ability Extraordinary, the Era of \"Software Generated on Demand\" Has Arrived\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5's coding ability is extraordinary. In multiple coding benchmark tests, GPT-5 outperforms any other model. It achieved 42% in the final human test and even reached 75% in the SWE benchmark test.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">During the launch event, Yann Dubois, the head of later training at OpenAI, demonstrated GPT-5's powerful coding ability live on stage, asking it to generate a website for learning French with interactive games. GPT-5 wrote hundreds of lines of code in just a few seconds and directly displayed the front-end interface of the website.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In addition, GPT-5 also demonstrated a 3D game created with just one prompt, which has exquisite graphics and very realistic physics effects.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">More Secure, More \"Honest\"\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">OpenAI tested GPT-5 for over five thousand hours to ensure its safety. GPT-5 performs better in handling multi-step tasks, and will not \"confidently lie\" as before. For some issues that may involve security risks, GPT-5 uses a \"safe completion\" mechanism, providing as helpful an answer as possible while maintaining safety constraints.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">Fully Open, Multiple Versions to Meet Different Needs\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5 is open to all users, including regular users, Plus members, and Pro members. Regular users can experience GPT-5 for free, but with limited usage; Plus members get more usage; Pro members can access the stronger reasoning capability version of GPT-5 Pro. The GPT-5 series also includes versions such as GPT-5-mini and GPT-5-nano to meet different user needs.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Regarding pricing, the standard version of GPT-5 costs $1.25 per million input tokens and $10 per million output tokens. The pricing for the mini and Nano versions is relatively lower.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cstrong style=\"font-size: 18px; color: rgb(255, 153, 0);\" class=\"ql-lineHeight-1-75\">Integrated Architecture, Intelligent Submodel Scheduling\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">GPT-5 adopts an embedded three-in-one integrated architecture, consisting of the GPT-5-main model for handling routine problems, the GPT-5-thinking deep thinking model for solving complex tasks, the routing mechanism for real-time decisions, and the mini version that starts after the quota is used up. The routing mechanism quickly decides which model to use based on the type of conversation, complexity, tool requirements, and the user's explicit intention, automatically matching the most suitable model without the need for manual switching.\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>\u003Cspan style=\"color: rgb(187, 187, 187);\">【News Source】Jinrongjie Finance \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002F%E6%8A%80%E6%9C%AF\u002F%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD\u002Fai%E9%A2%86%E5%9F%9F%E8%BF%8E%E6%9D%A5%E9%87%8D%E5%A4%A7%E7%AA%81%E7%A0%B4-gpt-5%E9%9C%87%E6%92%BC%E5%8F%91%E5%B8%83-%E6%80%A7%E8%83%BD%E5%85%A8%E9%9D%A2%E5%8D%87%E7%BA%A7-%E5%A4%9A%E9%A2%86%E5%9F%9F-%E5%B1%A0%E6%A6%9C-%E5%91%8A%E5%88%AB-%E5%B9%BB%E8%A7%89\u002Far-AA1K7ABJ?ocid=BingNewsSerp\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">http:\u002F\u002Fu5a.cn\u002FgVkVC\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（This article is reprinted by this site to provide readers with more information, and the content does not constitute investment or consumption advice. If there are any doubts about the facts in the article, please verify with the relevant party. The views expressed in the article are not the views of this site, and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A01381d04-878c-4559-92e7-b6d9075e99ef%3A0.wav?Expires=1774838500&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=jWDnDwZr3xpWFacHNJ7PryQIWNA%3D","01381d04-878c-4559-92e7-b6d9075e99ef",8957912]