Sam Altman (山姆·奧特曼)

Codex can now work more persistently towards your goals, whether it's for hours or even days. The goal functionality is now available in the Codex app, IDE extension, and the CLI. With it, you can specify a concrete goal to achieve, and Codex will continue to work until that goal has been met. Let me show you how it works. Start by typing /goal in your message composer to enable the goal mode. Fro

Workspace agents in ChatGPT help teams turn repeatable workflows into shared agents that can pull in context, use tools, and move work forward on their own. In this demo, you’ll see how admins and builders can set safeguards for workspace agents. Enterprise admins can centrally manage who can use, build, and publish agents, and define what agents are allowed to do. Builders can also configure when

ChatGPT for PowerPoint changes the way you build presentations. It turns scattered company context into a streamlined workflow, helping you create an exec-ready, customer-facing presentation in a fraction of the time. When I open the ChatGPT window, I can start by organizing my thinking before I build a single slide. I can bring in context from connected apps like Google Drive, Microsoft 365, and

Appshots are a new way to bring the context of what you’re working on into Codex. On your Mac, press Command-Command to attach your app window to a Codex thread. Codex gets both a screenshot and text from the window, including content beyond what’s visible onscreen. Appshots are available across plans on Mac, with enterprise access coming soon.

With GPT 5.5, we see an increase in the number of facts that we can pull directly from the provider and patient conversation, and that directly impacts the quality of the note that we can deliver to give providers the best possible experience for their patients. I'm Matt Sanders. I'm the engineering manager of per note generation at Abridge. We come together in the conversation between the provide

Codex wrote majority of AI defense and every new feature that we're building is 100% written by Codex. We built AI defense where every single one of our engineer was equipped with Codex. The features that we were working on for AI defense would have taken several quarters for us to be able to get out in the hands of our customers and that dropped down to weeks. We built out this open source tool c

Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously

Codex is rolling out in preview in the ChatGPT mobile app on iOS and Android. In this demo, see how you can start, steer, unblock, and review Codex work from your phone while Codex keeps running on your computer, Mac mini, devbox, or managed remote environment, with your files and project context still in place. You can review outputs, approve commands, check progress, and keep work moving when yo

Financial models often lead to important decisions before they've even been fully reviewed. An Excel workbook containing a financial model might support a forecast, a board meeting, or other important financial decisions, but the very workbook may contain overwritten formulas, or broken cells, or even outdated data, which can lead to inaccurate assumptions about the financial model itself. With Ch

People are generating over 1.5 billion images a week in ChatGPT. In this episode, Product lead Adele Li and researcher Kenji Hata share some of the new use cases and trends since the launch of Images 2.0. Together with host Andrew Mayne, they trace the progress from the early DALL-E days and dive into the latest capabilities, including better text rendering, photorealism, multilingual support, wor

What Codex has really helped us do [music] is have small teams of people deliver massive value in in a very very condensed time frame. The shift has gone from us producing a lot of the code ourselves to us now overseeing the work that Codex can [music] produce and the quality of that has just kind of gone up exponentially.

Cortex can now help with almost everything. But only if it can work where actual work happens. And for many of us, that place is Chrome. To help with that, we just launched the Chrome extension for Cortex. It works with the Cortex app in both Windows and Mac OS. Let me show you what you can actually do with it and how it works with plugins and other ways that you can connect Cortex to the tools th

We’re introducing three audio models in the API that unlock a new class of voice apps for developers. With these models, developers can build voice experiences that feel more natural, respond more intelligently, and take action in real time: • GPT‑Realtime‑2, our first voice model with GPT‑5‑class reasoning that can handle harder requests and carry the conversation forward naturally. • GPT‑Realtim

When I saw the eval results come back, I said, "Is this true?" It was such a big leap from what we've seen in the past. Like, I literally did a double take. GPT 5.5 did extraordin difficult knowledge use case for finance being able to actually do a financial model projection based on structured and unstructured data. And this resulted in a 19 percentage point uplift from the prior version.

Ask ChatGPT to analyze, update, and explain your spreadsheets directly in Excel and Google Sheets. These add-ins are now available globally across all plans. Powered by GPT-5.5, ChatGPT helps you update models, analyze data, audit outputs, and turn messy business context into decisions faster. Now available globally for all plans. Get started: https://chatgpt.com/apps/spreadsheets/

When I saw the eval results come back, I said, "Is this true?" Uh because it was such a big leap from what we've seen in the past. Like I literally did a double take. GPD 5.5 is an absolute gamecher for our customers specifically in finance because the model is just amazing at multi-step reasoning. GPT 5.5 did extraordinarily well in a hard difficult knowledge use case for finance. being able to a

All right, so we're going to begin by asking Codex to look at my Google Calendar, because we know that we have a customer meeting coming up. It's going to ask it to list the attendees, the agenda listed, and it's going to go kind of swim through my calendar and pull back any information that seems relevant. Looks like we got our agenda and what to be ready for here. The next thing we're going to d

Hey everybody, I'm Pat Dennis and in the next minute I'm going to show you how to set up Codex. The first thing I do is I tell it what kind of work I do. This allows it to give me a personalized experience. #2: there might be projects that I was working on and other productivity or coding tools. I want to port those over into Codex with a click of a button. #3: I'm going to set up my plugins. I'm

"Codex with GPT-5.5 is currently state-of-the-art.” Arnav Singhvi dropped by to tell us how GPT-5.5 has been transformational for knowledge lift. “It's like a step size function change. Very good at handling multi-step workflows, agentic workflows." And because we love getting a Gold Star: OfficeQA Evals -GPT-5.5 in Codex had a 46% reduction in errors. This shows that gains aren't just theoretical

"Codex with GPT-5.5 is currently state-of-the-art.” Arnav Singhvi dropped by to tell us how GPT-5.5 has been transformational for knowledge lift. “It's like a step size function change. Very good at handling multi-step workflows, agentic workflows. And because we love getting a Gold Star: OfficeQA Evals - GPT-5.5 in Codex had a 46% reduction in errors. This shows that gains aren't just theoretical

What Codex has really helped us do is have small teams of people deliver massive value in in a very very condensed time frame. I'm Jordan Levy, the regional CTO at Andava, and this is what Codex unlocks. What Codex has truly [music] unlocked is two ends of the spectrum. One is senior architects like myself coming from uh complex environments. We're able to articulate what we want and Codex is able

Math is one of the clearest ways to see how far AI has come in a short span. OpenAI researchers Sébastien Bubeck and Ernest Ryu join host Andrew Mayne to explain what changed and what it could mean for the future of research. They reflect on how Ernest used ChatGPT to help solve a 42-year-old open problem, the difference between deep literature search and original mathematical discovery, and what

GPT5.5 is very precise and very token efficient. One of the things I've I've been very impressed is I've been meaning to create this internal tool that quite some some time and I've been like always like deferring it because I thought like it's going to take me days, but then yeah, I took Codex and I put the new GPT5.5 into it and then yeah, I was able to do it probably under 1 hour. One of the bi

[music] >> Hi, I'm Harsha and I'm going to show you how you can build a simple reporting agent [music] that helps create a regular weekly readout for your team. I'll start by adding a connection to the data source, which in this case is Google [music] Drive. This matters because the agent can work with the files, spreadsheets, where the data lives instead of someone manually moving information aro

Hi, [music] my name is Arjun. In this video, I'll be walking you through how we built Slate, our software review agent. [music] Slate follows the team's review process every time with its skill. [music] Skills define best practices and contain the necessary instructions for an agent to perform its work accurately and consistently. Slate [music] utilizes a skill to perform web research, checks appr

Hi, my name is Hojun [music] and I'll be showing you how to build Trove, a third-party risk manager which is similar to an agent our finance team uses [music] today to accelerate time-intensive vendor due diligence with more consistency [music] and control. We'll kick off the build process with a prompt which describes the workflow and the tools, skills, and systems [music] that this agent will ne

I think the magic moment was just the fact that it had enough intelligence to come back to me with a solution based on like this very abstract question that I asked it. >> [music] >> Hi, my name is Shaurya Joshi. I'm an AI researcher at Nvidia and I'm really excited for GPT 5.5 because I'm about to just go get a lunch and it's refactoring my entire code base for me. One of the things that I found

(0:00) Bestie intros (3:05) OpenAI misses targets, Codex gains on Claude (20:02) AI cybersecurity: a market that's about to explode (31:03) Elon vs Sam Altman lawsuit (41:00) Big tech smashes earnings, Capex explosion (52:44) Vibecoding nightmare: AI deleted someone's codebase (58:33) Retatrutide craze: peptides go mainstream (1:06:34) Friedberg's Supreme Court experience Apply for Summit 2026: ht

I'm Claire Vo. I'm the founder of ChatPRD and the host of How I AI. So Claire, thank you so much. You were one of the very first users to have your hands on GPT 5.5. What was your first impression? My first impression was time to cook, which is I just started shipping a bunch of stuff. I threw it, turned it on in Codex, and across many, many projects, as I was showing you, you were looking at it.

My first impression of GBT 5.5 is that it is it is different. Before previously a lot of my prompts had to be very detailed or very instructiony kind of where I'm trying to tell it like hey look in this part of the codebase do this u whereas with GPD 5.5 sometimes I become lazy and I kind of give it a very amigious task but then it will figure it out. It actually re it directs his research and exp