Logan Kilpatrick

So I'm excited for that just because the models are great at tool calling.

They have good general intelligence, can do a bunch of this stuff.

multimodal i think continues to be just like a theme and a trend that works really well for gemini um it's also something that inherently has lots of scale by default if you for example like everyone has million you know maybe not millions of pictures everyone has many thousands of pictures most of which they never look at um there's like many millions of pictures and video and audio clips and all this other stuff around the internet and in your sort of like work corpus and all these other things so i think those use cases like it historically um

It's actually where we saw a lot of initial Flash traction for previous iterations of Flash was like, how do I understand all this multimodal data that my company has or that I personally have and then actually have a system that I can take action or make use of that data where applicable?

So I think this is a continuation of that story.

And I think coding, both of you mentioned this already, but I think coding is...

one of these use cases where it really is you can sort of feel how good this model is it's better on sweet bench than uh which the caveat of that story is like sweet bench measures a very very specific type of coding if you've ever like looked under the hood of what sweet bench is doing it's maybe doesn't translate towards like what my personal coding workflows are what other people's might be um

But you see this for vibe coding, which I'm really excited about.

I tweeted the other day, I think 3Flash is the default model for vibe coding because historically the cost was actually a real limiter.

If you look at a lot of products, it was like, hey, you can use it once and then you have to pay because it's expensive to have great frontier coding models.

And I feel like Flash is challenging that idea, which is exciting because it means more people are going to be able to build stuff.

Yeah, and stuff that's good.

Stuff that works.

Yeah, that's a good question.

Actually, so one of the interesting trends that I think we're going to see, and I saw a bunch of early examples of people talking about this online.

One of them was like, I saw somebody make the comment that they have historically always opted for the frontier intelligence model as their sort of daily driver model.

So whatever the most intelligent model was, was the one that they would go with.

And they were sort of willing to pay the sort of latency hit on that because they wanted frontier intelligence.

And they said that flash was the first model that has made them change that perspective because really being able to sort of have a model that's smart enough that you can sort of quickly that like iteration speed actually matters a lot.

And like it keeps you in flow state and makes it so that you're not, you know,