Jyunmi Hatcher

For developers and enterprises, Google is pairing the model with anti-gravity, a new agent-based coding environment where AI agents plan, write, run, and debug code across an integrated editor, terminal, and browser.

In context, this is Google's answers to a crowded field of so-called frontier models as it tries to stand out from OpenAI and Anthropic by shipping faster into everyday products instead of chasing leaderboard scores alone.

And those are also very impressive.

So Brian, I know you got a chance to deep dive, do some use cases and demos.

So tell us what you've seen in Gemini 3 so far.

Yeah, I mean, it makes sense.

But I think what it really highlights is that Gemini 3 really takes advantage and is aware of its entire context window, right?

It's that million, million tokens.

Yeah.

Right.

Oh, is it at two million?

So, I mean, that's that's a huge leap because that's always been an issue.

You could have a huge context window, but it was how much of it is it going to remember?

How much of it is it is it going to recognize?

And that seems like a perfect use case to to show off that that message.

lack of a better term right now, memory of its entire context.

Not to mention that it's 2 million tokens versus was it 400,000 or something like that for GPT-5.1 maybe?

Maybe I'm just thinking about 200,000 for the 200,000.

Yeah, absolutely.

I do that.