3:14
Good morning and welcome to DevDay. We are glad to see everyone present here and those who joined us online. Thank you for coming. I think this DevDay will be our best event. You asked us to add Code X support for Linux. So, we did it. You asked. Yes. Wow, a lot of people like this. And you also asked for Code X for your phones. So we added it there too. We take great pleasure in creating things that are truly useful for you. And the next new feature is what you asked for the most. You've probably seen the first signs of what this might look like in virtual assistants helping you book a table or buy a plane ticket. These are useful features, but they are quite primitive compared to what the technology is capable of.
Therefore, we want to achieve much more. Dots are extremely capable and always-on agents, built to perform any task you can imagine. They introduce AI into a completely new format of work, freeing up your time and attention for truly important tasks. What's new about my "dot" is that over time it learns how I work and what I want to focus on. Sure, Dots can handle small tasks like booking tickets or choosing a dinner time, or helping with that renovation you've been putting off. But since Dots is based on Astra and our other developments, you can delegate serious tasks to them.
Port our application from this legacy API until it is disabled. This is a complex problem. The API can be used throughout the codebase, and it is not always obvious what will break after making changes. Dots get access to everything they need to work. They respect your access rights. They have their own computer in the cloud, their own browser, and they can write and test code as needed. You also don't need to be in ChatGPT to chat with your dot. Soon you will be able to send him text messages. He will respond to you by text. It can send you updates, notify you of changes, or ask for your opinion on a decision. You can even call your dot and talk to him directly. Dots also work with over 4,000 apps in
ChatGPT. When you connect your dot, it accesses the same plugins you already use. He can simply follow your work wherever it takes place. Dots are powered by Astra, our most consistent model, and have built-in protection and security features. You can set limits on app and computer usage, and set instructions for any action your Dot takes. We present ChatGPT Space. This is a space where you, your colleagues, and your agents can work together. Space starts with a page similar to the one you see on your screen. You can write a plan, conduct research, create an image, generate data, or create something new. Dots can work with you directly on this page. You can bring in Dot and ask him to track dependencies, just like you would with a senior engineer. You
can tag Dot in a comment and he'll get to work right away. But the spaces seem alive. You can give the page specific instructions, such as: " Check the API channel in Slack and update the results here daily." We'll also soon introduce a new way to create presentations right in Space. So you can share a presentation with your team and work together with your Dot and their Dots right in ChatGPT. So, today I'm going to show you how the teams at OpenAI are using Dots on the example of a fictional music app called Blossom. This is a fictional cool app for creating playlists, and this is my Dot-Dotty. It's a second pair of eyes for everything I give it access to. And, more importantly, a second
pair of hands to really take some of the work off my shoulders. This morning he thoughtfully highlighted a bunch of things I should pay attention to. Our launch review was moved up earlier and he has already updated my calendar. We released new features to test users last night, and he's already reviewed the feedback. It looks like the designers made a last-minute change to our home screen, and Dotty not only noticed it, but sent me the design so I could take a look right here. So, let's take a look at this new design. Okay, I like it, but before the launch I want to try how it works, right? Since Dottie has access to this laptop and its development tools, I'll just say, "Go ahead, create." And no pressure, Dottie. We'll check that
later. And since Dottie already knows who I am, what I'm working on, and why, I can get straight to the point. It's like talking to a second brain. Let's say I walk into the office and I need to catch up on yesterday's alpha changes Hello, I'm just coming in. Um, can you fill me in on yesterday's Hey, I'll discuss this with the go-to- market leaders later. Do you want to just add the FAQ to the page?" Of course, I'll take care of that Okay, thank you. We've added a new place to collaborate with your team in ChatGPT called "Spaces". When I press the slash, you see everything that can be created., tables, live prototypes-everything you need to bring your ideas to life. Your team and your " dots" can join you here. Let me show you what I mean. I can
tag colleagues like Rennie or Rishabh, but I can also tag Dottie herself and say, "Hey, can you turn this into a bar chart, please?" The visualizations here are fully dynamic. Here's the FAQ we just asked Dottie to add. And since Dottie has all my context, I can just delegate to her what's slowing me down. For example: "Hey, let's add those adoption stats from the engineer from my Slack messages." And Dottie can find her. Or something like, " Hey, let's take more feedback from Laura into account." But where Dottie really changed the way we work is in Slack.
We are all being tagged all day long. And sometimes an innocent request, like agreeing on launch specifications, turns into an entire project. Now I can forward the discussion to Dottie. I can just delegate it and say, "Hey, can you take this on?" As you can see, my messages are full of such small tasks. I decide what needs my attention and what my " dot" can handle. Here's something we at OpenAI didn't expect at all. People began to treat their "dots" as their representatives. Now we add our "dots" to group chats from the start. Here are Gabe and Patrick delegating tasks to their " dots." When the "dots " return with the results, everyone stays up to date. And our engineers went even further. Here someone sent the session ID and
user error to the feedback channel. Gabe's "dot" picks up on this, immediately investigates and opens a PR with a fix. And this is a real example. Our engineers fix dozens of bugs every day with Dottie. It's amazing to see how much Dottie has created for Dottie. This is a live demo, and there seems to have been a small error in this Codex thread earlier. So I'll try again. And what happens now is that Dottie can use Codex on my laptop to build an app and run it in the iPhone simulator. This may take some time. But the most amazing thing is that we can start with a mockup on my phone, run local execution, and turn that into a working app. And then I can
ask Dottie to just keep working and send the PR while I'm here with you. Both Dottie and Space will be available to all ChatGPT Pro, Business, Premium, and Enterprise users starting today. Dottie is included in your plan, and calls don't count against your usage limits. You can always work with your Dottie. It is available 24/7. Today we are also introducing specialized versions of Dottie. These are virtual colleagues set up by your company who will work alongside your entire team. They will help scale labor-intensive work like accounting, marketing, law, and more. You can set goals and provide context, review results, and provide feedback for improvement-all available to your entire company. We tested them internally at OpenAI, and they proved to be
extremely effective in many different roles. So we invite our corporate clients to collaborate to customize them in the best possible way. We are also working with Microsoft to integrate specialized Dottie into Agent 365. This means that companies will be able to manage their Dottie through the Microsoft tools they already use. We are launching GPT 6.1 Soul. You don't even know what it is yet. It provides near Astra-level intelligence at a fifth of the price to help you program and create. It is extremely accessible to use. In fact, data entry costs are 95% off compared to standard ones. This makes it much cheaper for agents to reuse context, iterate decisions, and complete longer tasks. We are sure you will love this
model, which is a bit smarter than the Astra, and it will become your daily work tool. So we're really excited to launch 6.1 Soul today. We run Ultrafast via API, ChatGPT, and Codex. Fast already gives you double the speed for double the price of standard, but Ultrafast delivers eight times the speed for six times the price of standard. An incredible 300 tokens per second. You know what? It's worth it. Um, let me show you. So, it's the same task, relatively side by side. We gave both models the same request. Create and launch a rocket in DevDay colors. On your left is a superfast model. On the right is the usual one. You see how much faster
it builds. The superfast rocket is already taking off, but the regular one is not even close to being ready. There is nothing like it on the market today. You can use the super-fast model with Astra now, and soon we will offer it for 6.1 Soul. The solutions API allows the model to respond in a split second. It's so fast. So the interaction feels almost instantaneous. You see here how she uses the computer and how quickly she goes through these steps. This is not a sped-up video. This works by giving our Luna model a set of predefined options: routing a request, classifying an image, or deciding what the agent should do next.
By focusing the model on this choice, we make it extremely fast while maintaining capabilities such as image understanding, broad language support, and security features. It all starts with the Codex framework. It powers Codex, Work, and now Dots. We have put a lot of effort into making this framework as efficient as possible, so that you get accurate results with the highest speed and the minimum number of tokens. We make it all available in the cloud. This is what I know you've all been waiting for. Codex is now completely in the cloud. It's been incredible watching you create 3D worlds with Astra over the past few weeks. I took some regular photos of the place to recreate the same thing here. So, let's take
a look. What's interesting here is that I'm going to interact with this room using a superfast model. So what I'm going to do is this: if we have a little difficulty with voice chat, I'll come back here and we'll do something else with this room now. It's a little empty here, so let's say, "Why do n't you put dots in places, for example?" That would be good, because it's too empty here right now. Time to draw the dots, and here they are. We are simply changing this world in real time as we work. Pretty cool. Maybe we should try something more risky now? Why don't we just turn the camera to the hall, to the audience, and I'll say: put the live broadcast on the big screen. And let's see what comes of it. Wait, like this. Oho. So if you want, wave to
the audience. To those watching us on YouTube, thank you for joining us. We've completely updated the Kodak CLI, adding a bunch of improvements you've asked for. So, let's start with this empty React project. You know what? We'll still do it ultra-quickly, but this time simply through the command line. So we'll say: create an app that randomly selects three participants. I want to give them free tickets for next year. Let's see what comes of this. Of course, it would be even faster and better by voice, but we implemented two-way audio communication in the terminal. I would like to show you this, but it all works on GPT live, just like in the application. This is not ordinary dictation, you can react in real time. A few seconds- and boom, our application is
ready, ultra-fast. Make six participants. Why not? Don't blink, it has to be quick-and boom, it's already changed. Yes? So, okay, let's expand this to full screen. Shall we click this button? Okay, let's do this. Let's see who will now return for DevDay 2027. Six winners, congratulations. My bot will take care of it and send you the details later. Show us how you play the game. The interesting thing is that now we are creating a connection through a browser. This means that Astra will understand how to control the game and learn to play it in real time. So in a second she can start the game,
choose a ship on her own, and if you look at the screen on the left, you can see the decisions the model is making. I love watching Astra figure out a game designed for us humans and learn how to play it. Cool, right? Finally, I want to show you another demonstration that combines all of these modalities. Our friends at Hugging Face have kindly provided us with an early version of their new programmable robot, a micro-duck, and her name is Lavender. Last attempt, I asked my colleagues last night to connect everything here, so let's see if we can connect. Look at the audience and draw a cool picture.
Okay, you're in this picture. Let's see what happens. Hey, Lavender, can you croak if you can hear me? Are you excited that we're at DevDay? Excited DevDay. And here's a photo of all of you that was just taken. Finally, to put it all together, we want to make it easier for you to sell products to more companies. So today we are launching the OpenAI marketplace. Today, we have over 30 partners, including Code Rabbit, Notion, Vercel, and others. And our corporate customers can now use a portion of their OpenAI commitments to purchase your products. This gives you another way to engage clients who already work with
us. And, of course, if you have obligations to OpenAI yourself, you can spend them on the marketplace as well. Thanks to our partnership with Base 10, you can now access open source models on the marketplace. So, thank you. We released a lot of new stuff this morning. New ways to work with artificial intelligence using dots and space. New models with GPT 6.1 Soul and super speed. Solution APIs, agent APIs, Codex in the cloud-many new ways to help you build your business, attract customers, and more. This is a big day for us, and we look forward to seeing what you do with it all.
