GPT-5.4

All right? And expert judges then compared the unlabeled AI and human outputs in a blind head-to-head pairwise comparison.

And what we've seen now is, as an example, the best model for actually creating front-to-back economically valuable outputs like a human would is the OpenAI's GPT 5.4 model, and it matches or exceeds industry professionals in eighty-three percent of these evaluations, right? I remember when some of these first GDPVal numbers came out, and I'm like, "Oh, that's, that's pretty impressive," right? When we were in the thirty, forty percent.

But now it's, it's undeniable.

And I do assume that by the end of the year, that number is gonna be like in the mid-nineties, right? My-- Like, one of my predictions is it was gonna get to eighty percent, and we're already at eighty percent.

Whereas obviously, uh, you know, Codex, uh, by OpenAI and Claude Code by Anthropic run those respective models.

So, uh, like I mentioned, uh, Cursor did hit two billion ARR in February of this year with over one million paying customers, and then you can use the best models, right? So you can use OpenAI's GPT 5.4, you can use Claude Opus 4.6, Gemini 3.1, or Cursor's own Composer model.

Next, Windsurf.

So Windsurf went through kind of a, a weird middle of twenty twenty-five.

enough? I personally have been extremely surprised with what the models are able to accomplish.

I'm sure Phil can talk about this as well, but there's many things that I've seen it do that As we mentioned, back with the GPT-5.4, kind of that inflection point where, I mean, now you look at some of the things, some of the just fascinating and creative solutions that it has to problems, where it really kind of makes you think, it's like, wow, that's extremely impressive that it was able to come up with that.

Can you give me an example?

Yeah, can you think of something off the top of your head? I mean, some of the solutions that it has in kernels, some of, again, going back to the heuristics things that we had talked about where it's able to determine how to best balance requests across accelerators.

It didn't refuse doing it.

GPT 5.4 managed about 33 times.

And a year earlier though, Opus 4 was around 6% and GPT 5 was zero.

Think about that trajectory.

They're wrapping Anthropic's APIs.

They're giving you versions of GPT 5.4. codex etc they're giving you versions opus 4.5 at all the different variations of it and they're sprinkling their own abilities on top of that and amp is i don't know if it's source graph because that's where its roots came from but um what makes it so good i'd love to like learn what makes it so good but amp i have you played with amp by any chance

here i have not no

Well, after this podcast, go and play with AMP.

Fascinating.

OpenAI's August 30, 2026 release notes confirm the termination of the official DLE GPT alongside the deployment of GPT 5.4 Mini for free-and-go users via the thinking menu.

How easily will everyday users adapt when explicit legacy creation tools are quietly replaced by these highly integrated, compact background models?

Well, the operating system now functions as an invisible router, analyzing your request in real time and autonomously deciding whether to deploy a text model or a data parsing agent without requiring explicit tool selection.