内容来源：https://www.wired.com/story/openai-dev-day-sam-altman-chatgpt-apps/

内容总结：

在周一举行的年度开发者大会上，OpenAI宣布推出全新第三方应用嵌入功能，允许开发者直接在ChatGPT环境中集成各类应用程序。公司首席执行官萨姆·奥尔特曼在旧金山会场宣布，这一变革将“催生具备自适应、交互性和个性化特征的新一代应用生态”。

目前部分开发者已可基于开放标准，通过预览版应用软件开发工具包在ChatGPT内构建应用。初期应用分发权限仅向少数大型合作伙伴开放。现场演示环节中，OpenAI软件工程师亚历克西·克里斯塔基斯通过对话指令，先后调取Canva设计平台生成宠物看护业务宣传海报与融资路演材料，继而结合Zillow房产应用在匹兹堡地区筛选带庭院的三居室房源，全程通过自然语言交互完成复杂操作。

奥尔特曼透露将推出“智能商务协议”实现聊天界面即时结算，为开发者开辟商业化路径。此举标志着OpenAI从聊天工具向“超级助手”战略转型的深化。公司产品负责人尼克·特利在会后简报中表示，ChatGPT旨在构建集成立业生态的智能中枢，而非局限于对话功能。

为巩固开发者生态，OpenAI同步发布多款工具：代码生成模型Codex结束测试全面开放，新增Slack代码编辑功能与使用分析系统；推出可视化AI工具开发平台AgentKit；时隔数年再度开源GPT-OSS模型，并通过API发布新一代旗舰模型GPT-5。这些举措被视为应对Meta、DeepSeek等企业开源模型竞争的关键布局。

在算力基建层面，OpenAI近期签署了价值60亿瓦的AMD芯片采购协议，或将持有该芯片制造商10%股份。奥尔特曼强调：“当前正是史上最佳创业时期，从创意到产品的转化速度已达到前所未有的水平。”

（本文根据科技记者里斯·罗杰斯现场报道整理）

中文翻译：

周一，OpenAI宣布推出一种将第三方应用直接嵌入ChatGPT的新方式。在旧金山举行的年度开发者大会上，首席执行官萨姆·奥尔特曼表示此举将“催生新一代能够对话的适应性、交互式个性化应用”。

即日起，部分开发者将能通过预览版应用软件开发工具包，基于开放标准在ChatGPT内构建应用。目前仅限少数大型合作伙伴具备分发这些应用的权限。奥尔特曼现场演示了Spotify、Canva和Zillow等应用在对话界面响应文字指令的运作场景。

演示环节中，OpenAI软件工程师亚历克西·克里斯塔基斯向Canva应用发出指令，要求为宠物遛养业务设计宣传海报，随后又令其生成项目融资路演稿。接着他让ChatGPT推荐业务拓展城市（系统建议匹兹堡），并调取Zillow应用展示当地待售房源。交互地图呈现结果后，克里斯塔基斯进一步要求筛选带庭院的三居室住宅。

奥尔特曼透露，OpenAI正规划开发者变现方案，包括通过聊天机器人实现交易功能。“我们将推出智能商务协议，支持在ChatGPT内即时结算。”这并非OpenAI首次布局应用生态。两年前开发者大会曾发布自定义组件GPTs，截至2024年1月GPT商店上线时已积累300万组件，但市场反响未达预期。

对于与Canva、Zillow、Spotify等合作方的收益分成细节，OpenAI未予披露。新版SDK的发布彰显其深化企业合作、构建用户闭环的决心。若将过去三十年的网络与移动时代定义为浏览网页与沉浸应用的体验，OpenAI正试图打造聊天驱动的融合操作系统。

ChatGPT产品负责人尼克·特利在主题演讲后表示：“我们的初衷并非打造聊天机器人，而是构建超级助手，过程中略有偏离。”他指出公司在自然语言处理领域的突破令人振奋，但这家估值5000亿美元的初创企业将持续探索多样化交互界面。

“用户不会全天候使用ChatGPT，但可能以它开启每一天，再转向其他应用。”特利补充道。除重构现有应用外，OpenAI致力于成为AI智能体开发的核心推动者，最新发布了包含可视化构建工具AgentKit在内的多款开发套件。

争夺开发者生态的关键战场同样聚焦编程工具。本次活动宣布代码生成模型Codex结束研究预览全面开放，同步推出代码问答、Slack即时编辑、SDK及企业用量监控等新功能。面对谷歌、Anthropic和亚马逊的竞争压力，以及Meta、DeepSeek等开源模型的冲击，OpenAI在八月发布了数年首款开源模型GPT-OSS，并通过常规API推出万众期待的GPT-5旗舰模型。九月其针对Anthropic和谷歌的编程模型升级，以及三月发布的智能体开发工具，均显现出激烈攻防态势。

去年开发者大会曾公布低延迟语音交互API、视觉模型微调方案及大模型驱动的小模型优化工具。在生成式AI经济可行性备受质疑之际，OpenAI正大规模建设数据中心。今晨宣布收购AMD价值60亿瓦特芯片的协议，或伴随获得该芯片制造商10%股权。

奥尔特曼在主题演讲中强调：“此刻是史上最佳的创造者时代，从构想到产品的转化速度空前提升，这种加速态势切实可感。”

里丝·罗杰斯对本文亦有贡献。

英文来源：

On Monday, OpenAI unveiled a new way to embed third-party apps directly into ChatGPT. At the company’s annual developer conference in San Francisco, CEO Sam Altman said the move would “enable a new generation of apps that are adaptive, interactive, and personalized, that you can chat with.”
Starting today, some developers will be able to use a preview version of a new apps software development kit (SDK) to build apps within ChatGPT using open standards. The ability to distribute these apps is currently limited to a handful of big partners. Altman showed off several examples of how these apps would ultimately work within ChatGPT. The demo featured Spotify, Canva, and Zillow apps appearing inside a chat and responding to typed commands.
In one example, OpenAI software engineer Alexi Christakis launched a chat directed at the Canva app and prompted Canva to draw up posters for a dog-walking business. From there, he asked for a pitch deck in order to raise capital for the business.
Christakis then asked ChatGPT to suggest a city that would be ideal to expand the business into (it offered Pittsburgh, for those wondering). Next he called up a Zillow ChatGPT app and asked it to show homes for sale in the city. This brought up an interactive map showing houses for sale. Christakis asked ChatGPT to refine the results to just three-bedroom houses with a yard.
Altman said that OpenAI plans to introduce new ways for developers to monetize their ChatGPT apps, including methods for buying things through the chatbot. “Soon we’ll offer an agentic commerce protocol, with instant checkout from right within ChatGPT,” Altman said.
This is not OpenAI’s first effort to introduce ChatGPT apps. The company announced a way to build custom widgets or GPTs at its developer conference two years ago. When the GPT Store officially launched in January 2024, OpenAI said that developers had created over 3 million custom GPTs. Ultimately, however, the widgets did not prove a big hit.
OpenAI didn’t share details around any revenue-share agreements with Canva, Zillow, Spotify, and the other apps it highlighted today.
The new SDK announcement signals a deeper commitment to working with established enterprises and app makers—and an emphasis on keeping users within ChatGPT itself. If the web and mobile eras of the past 30 years were defined by users browsing the web or being locked into a mobile app experience, OpenAI is now combining the two into its own kind of chat-driven operating system.
Nick Turley, OpenAI’s head of product for ChatGPT, said in a briefing after the keynote that the company “never meant to build a chatbot; we meant to build a super assistant, and we got a little sidetracked.” He indicated that OpenAI is most excited about what it has achieved in natural language processing, but that the $500 billion startup will continue to experiment with different user interfaces around that.
“Will people spend all of their time in ChatGPT? I don’t think so,” Turley said. “I can imagine you starting your day with ChatGPT,” then being guided toward other apps and websites.
Beyond reimagining existing apps, OpenAI hopes to put itself at the center of efforts to build agents that use AI to complete tasks on a user’s behalf. The company unveiled several tools for building agents including AgentKit, a drag-and-drop interface for building advanced AI tools.
Capturing developer mind-share is also, of course, about coding tools. At Monday’s event, OpenAI announced that Codex, a model optimized to write code, would come out of research preview and become generally available. The company also announced new Codex tools, including a way to ask questions about code and edit it via Slack messages, an SDK for the Codex model, and new analytics tools to allow companies to monitor their employees’ Codex usage.
OpenAI is competing with Google, Anthropic, and Amazon in trying to lure developers to its services. The company is also under pressure as rivals like Meta and DeepSeek launch capable open source models, which allow developers to fine-tune premier models for their needs, without having to pay for access to OpenAI’s APIs.
This August OpenAI released its first open source model in several years, GPT-OSS, and also made a long-awaited new flagship model, GPT-5, available through its regular API. OpenAI also upgraded its coding model for developers in September, a move seemingly designed to respond to very capable coding models from both Anthropic and Google. OpenAI debuted tools for building agents in March.
At last year’s developer day, OpenAI announced an API for low-latency voice interactions, ways of fine-tuning vision models, and a tool for fine-tuning small models using OpenAI’s more powerful ones.
OpenAI is also building a vast empire of data centers at a time when questions have begun to swirl about the economic viability of generative AI. This morning, the company announced a major deal to acquire 6-gigawatts’ worth of AMD chips in a move that may also see the company take a 10-percent stake in the chipmaker.
During today’s keynote, Altman suggested that OpenAI was already helping software developers reach new levels of productivity—a fact that might help ramp up demand for all that compute.
“This is the best time in history to be a builder,” Altman said. “It has never been faster to go from idea to product. You can really feel the acceleration at this point.”
Reece Rogers contributed to this report.