Introducing GPT-6 Astra for developers

0:04GPT-6 Astra is here. It's our latest frontier model and the best model in the world for tasks where raw intelligence matters. For my own projects, Astra feels like working with an experienced collaborator. I'm able to hand it bigger, less well-defined tasks with minimal handholding. And whether you're coding, writing, designing, or just working on things that require a little bit more intellectual horsepower, you'll find that Astra delivers great results,

0:30often on the first attempt. So let's talk about the model's new capabilities, starting with computer use. You might already be familiar with computer use. It's a capability that lets a model use a computer or browser like you or I would. With Astra, the model is more accurate and more efficient when using a computer. It's easiest to show this in the ChatGPT app, but Astra's improved skills

0:51are also available in the API. I'm going to have it illustrate where I'm standing. I'll take a photo, then send it to Codex to paint. Use Krita to paint this picture in the style of Van Gogh and add the Golden Gate Bridge in the background. Astra has to keep track of the picture I asked for while working

1:09through the app's controls. It's using screenshots to see what's happening while keeping the app in the background, meaning I can stay focused on other work. The model is able to paint a hopefully recognizable picture far faster and far better than I could have done. The same ability can help with the workflows developers use every day, whether you're filling out forms for an API key or performing

1:29QA on a mobile app. When it comes to creativity and knowledge work, we've also seen Astra produce significantly better outputs than previous models. We took several prompts of demos, games, and tools we've built with previous models and rebuilt them at different reasoning levels. Across the board, Astra has more attention to detail, better understanding of the user's prompt,

1:49and can build more sophisticated outputs. In particular, it excels at building 3D models. I've seen it make incredible renderings of gardens, shipyards, animals, cityscapes, even Dyson spheres. And in my own experience, I've found Astra's outputs, from slideshows to essays, significantly easier to read and work with than other models I've tried. Its personality feels more natural to me, especially in the back and forth

2:13when I'm giving it feedback. As you might expect, Astra excels at long-running tasks. And as you give models longer, more involved pieces of work, you need better ways to work with them. That's why we're bringing asynchronous tool calling and steering to the Responses API. First, async tools. With asynchronous tool calling, the model can keep working on other parts

2:34of a task while a tool call runs. When the results are available, it can use that information to continue. Steering lets you give the model new context or change its direction while a response is still in progress. If you view steering in Codex, this should feel familiar. Let's go ahead and give the model a task and then change what

2:53we're asking for while it's working. You can see it pick up my new instruction and adjust its next steps while the tool it already started continues running. I can change the direction without having to cancel the running tool or start the task over. And I get the final result with the new context incorporated. Astra is now available in ChatGPT,

3:12Codex, and the API. Try it on your most challenging tasks and take a close look at the results. Happy building.