Kevin Rose

iFramely won the summary.

Gemini won the actual main content.

And then what I hit is I hit GPT-5-mini.

Largely because it's fast as hell and cheap, and it's quite smart.

Anthropic has a handful of models that also fall in this camp.

I mean, they all do.

It's really, at the end of the day, because I'm using vector embeddings from OpenAI, it's just when I have one model provider, I just stick with it, unless I really need to bounce around, which I did for the actual Gemini crawl and understanding of other things.

So I try to keep it as simple as possible, but sometimes you need multiple models, and I

I will say Vercel's AI Gateway is a great way to kind of code once and just flip a model on the go.

So I highly recommend checking out Vercel AI Gateway as a way to quickly swap models rather than having to recode it.

All right, so I want a TLDR.

That's important.

I just want that human readable so you can see that.

So I want a TLDR that is a vector TLDR.

And so for people that don't know, vector embeddings are really interesting.

I had never worked with them before.

I had heard about them.

I understood technically how they function.

But the point is that if you take a keyword-rich and kind of deep understanding piece of content, you can create mathematical representations of that content and embed these with OpenAI.

and store them in Postgres by using a vector extension.