Damn it, OpenAI really went crazy this DevDay, updating a lot in one breath. I have compiled the most important things from the early morning press conference into 7 items for everyone: 1. Dots: OpenAI's personal agent This should be the most important product in the entire venue. Each Dot has its own cloud computer, browser, and execution environment, which can help you search for information, operate web pages, process files, run tasks 24 hours a day, and even call the plugins already connected in ChatGPT. Simply put, it is OpenAI's official version of OpenClaw. In the future, you can even find your Dot to work directly through SMS or phone calls. 2. ChatGPT Space: A shared workspace between humans and agents Previously, ChatGPT was a single chat window. In the future, documents, images, data, plans, charts, and project materials can be directly placed in Space. You can either @ colleague or directly @ Dot: Update today's new data Agents will do it themselves. GPT-6.1 Sol: A Cheaper Agent Model GPT-6 Sol has only been released for a week, and 6.1 Sol is here again. The ability continues to improve, but the biggest focus is actually on reducing costs. Especially as cache prices continue to decline, computer operating capabilities have become very close to Astra, but the cost per task is only a small fraction of Astra's. In the future, if the agent runs tasks 24 hours a day, the cost will become lower and lower. 4. Decisions API: specifically designed for AI to do multiple-choice questions This is very interesting. Many agents don't need to write a long paragraph every time, they just need to decide: Pass or reject? Should we execute A or B? Which agent should I assign this task to? OpenAI now provides a dedicated Decisions API directly, allowing GPT-6 Luna to quickly make such judgments with sub second latency. Jev just made this direction popular two weeks ago, and OpenAI immediately followed suit. 5. Codex is fully cloud based Codex Cloud can now save reusable development environments. The code repository, dependencies, tools, and environment variables are pre configured so that the agent can start working directly without having to rebuild the environment every time. Your MacBook can even shut down directly, allowing Codex to continue writing code in the cloud. Codex Security has also been upgraded, allowing for scheduled scanning of code repositories, discovery of vulnerabilities, validation of issues, and preparation for fixes. More and more like a resident programmer. 6. ChatGPT plugin begins to become a real app Previously, plugins only called APIs in the background. Now applications like Figma and Adobe can directly use editors Dashboard、 Insert forms and workspaces into ChatGPT. In the future, there may be no need to jump out and open a bunch of websites. Many software will directly "live in ChatGPT". OpenAI has also launched a Marketplace to help third-party software directly reach enterprise customers. $7.500 Pro member has arrived The previously rumored $500/month package has officially been launched. Higher credit limit and Ultra Fast are also available. At most, it can make Codex run several times faster than normal speed. Of course, the speed of burning credit will also take off together So looking at the entire press conference, I think the most important change for OpenAI this time is that ChatGPT is transforming from a chatbot to: Model+Agent+Cloud Computer+Workspace+Plugin+Programming+Enterprise Software Market. In 2023, everyone opened ChatGPT to ask AI a question. What OpenAI wants to do in 2026 is for you to open ChatGPT and directly hand over the work to a group of agents. This is what makes Dots really interesting.