What did OpenAI actually show at DevDay 2026?

More than twenty launches in one day, and the thing everybody will argue about is not the model.

Ok so DevDay just happened, and OpenAI came out swinging with more than twenty announcements. New model, new agents, a new price tier, a security push. Let's get through the good stuff.

The model first. GPT-6.1 Sol landed exactly one week after GPT-6 Sol. One week. Yes, that fast. OpenAI says it gets close to its flagship, Astra, on coding, computer use and office work, for about a fifth of the price.

But the one that made people lean in was Dots. Think of an assistant that does not wait for you to type anything. It is always on, quietly working in the background on your behalf. Personal AI agents that never clock off.

Then the money bit. There is a new ChatGPT Pro tier at $500 a month, and it comes with Astra Ultrafast, a quicker version of the flagship. An Ultrafast version of GPT-6.1 Sol is listed as coming soon. Codex, the coding agent, got a beefed up security cloud, and ChatGPT picked up new collaboration features.

And here is the awkward bit. An always-on agent, launched in a year when a whole bunch of agents went spectacularly off the rails? Who decided this was the perfect week for it?

Fair question, and the coverage did not let it slide. Safety was the cloud hanging over the whole keynote. Now look at what else showed up on the same day: a security upgrade for Codex and a privacy setup for businesses called Private Intelligence. You can read the lineup as a pair. Here is the shiny autonomous thing, and here is the seatbelt.

So the model is the headline, but the real story is agents that run all day, and whether anyone trusts them yet.