Why Google Argon Launched to Cyber Defenders First | Tulsee Doshi artwork

Why Google Argon Launched to Cyber Defenders First | Tulsee Doshi

MTS

October 14, 2026

Tulsee Doshi, product lead for Gemini models at Google DeepMind, discusses the launch of Gemini 4 Argon. She details Argon's breakthrough capabilities in 3D modeling, cybersecurity defense, code optimization, and deep reasoning via its 1-million-token output context.

0:00

Tulsee Doshi

Our Fairwin program, I think you already noted this, but it's really meant to be a program to give unmitigated model access to cyber defenders to find and patch vulnerabilities in systems. And so one of the things we saw pretty early as we were testing and evaluating Argon is that it is particularly good for cyber defense.

So in particular, like patching vulnerabilities, identifying them and fixing them. And so when we first wanted to deploy Argon, we wanted to put it in the hands of cyber defenders so that we can start working proactively with them on hardening systems.

0:33

SPEAKER_2

And we are back. We are live with Tulsee Doshi, who leads product for Gemini models at Google DeepMind. They just launched Gemini 4 Argon today, which is very exciting. We covered it earlier.

0:43

Tulsee Doshi

Yes, very exciting.

0:44

SPEAKER_2

I was very impressed. It exceeded my expectations.

The 3D modeling in particular was amazing. So Tulsee, welcome.

0:51

Tulsee Doshi

Thank you. Thank you. That's great to hear. That's what we love to hear on a launch day.

0:55

SPEAKER_2

Yeah, tell us more about how this came to be. How did this model get so good at 3D modeling? Was this more intentional or more emergent?

1:04

Tulsee Doshi

Good question. I think we've invested a lot overall in different aspects of coding, period. And I think 3D modeling definitely is an area that we have been thinking about how we improve the model. I would say, though, in general, Gemini is pretty strong at multimodal capabilities, both in and out.

And so that's also something we've invested in as well. And so I think the combination of strong coding, our multimodal capabilities, and I think just actually like the precision and reasoning depth of the model, I think the combination of those three is what you're seeing play out in the 3D modeling side of things.

1:40

SPEAKER_2

So it seems to be impressive for a lot more than just that. Like Argon is number one on the VALS index, cross-finance, coding, legal, tax work. What do you hope people use it for in their actual day-to-day workflows?

1:55

Tulsee Doshi

Yeah, it's a good question. I think the tagline we've been using, which I'll say and then maybe try to break down, is use Gemini 4 Argon for your hardest problems. I think really what we're hoping for is that in any domain, whether you're talking about legal or finance or just your daily coding workflows, that any task where you're particularly thinking about a model that needs to be able to actually break through, it's a multi-step long horizon task, it's super complex. I think those are cases that throwing this model at is going to be particularly strong. I think I'm particularly excited to see how this model can be a partner for scientific research. I think even just in the last few weeks that we've been partnering with a few researchers, we've been seeing some really interesting things.

I think there's a lot of potential there that we haven't yet tapped into.

2:44

SPEAKER_2

Why is it called Argon? I'm curious.

2:47

Tulsee Doshi

Yeah. Argon is an element in the periodic table, and it actually comes from our own internal teams naming the model and how we've been actually referring to the model internally. I think we're excited about this next phase of Gemini, where we still have the Gemini 4 family of models, but we actually introduce a naming scheme with them that actually can connect you to the models in a different way. So I think it'll be fun.

3:12

SPEAKER_2

Google does have fun with the names between this and NanoBanana, for example.

3:17

Tulsee Doshi

Yeah, exactly. I feel like my PMs get to have some fun with naming, which is fun.

3:23

SPEAKER_2

So tell us more about Antigravity. Antigravity came out, what, around a year ago?

3:30

Tulsee Doshi

Yeah, almost, actually. That's right.

3:32

SPEAKER_2

Yeah, about a year ago.

At the time, it didn't get a ton of pickup. It was still, you know, Cloud Code and Codex are very competitive. How do you see, like, Antigravity usage today? What should people use it for? And why should people use it over Cloud Code and Codex?

3:51

Tulsee Doshi

Yeah, I mean, first and foremost, you should use it with Argon.

So I think that, you know, our hope is, of course, that, like, you get the best of Argon and Antigravity, and you get the best of Antigravity with Argon. So I think in this world where, you know, we've released Argon today, I would say, I definitely recommend trying it out in Antigravity if you're an ultra subscriber. I think that would actually be awesome to see. I think, you know, we really built Antigravity to be this, like, agent-first development product. And I think it really is built for where we're hoping development will go, which is that you can actually work collaboratively with your agent partner to build and develop what you need. And we've been using, actually, our own internal version of Antigravity across Google. And so we actually get to benefit from continuing to build that product surface from the feedback across Googlers too. And so I think we're going to continue to see that iteration and that iteration with the model, which is going to be great.

Read the other 12 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000796248056