Home › AI news

Google's most powerful AI yet is here. You just can't use it.

Update, Oct. 14: Argon is now open to paying Gemini subscribers. Read the follow-up.

At the end of September, Google announced Gemini 4 Argon, which it calls its strongest AI model so far. Google says it's built for serious coding, business work, and cybersecurity, including finding and patching security holes in software on its own.

The catch is that almost nobody could use it at first. It was limited to the U.S. government and a handpicked group of cybersecurity teams. Google said paying business customers and its top subscription tier, about $100 a month, were next, with no date for free and regular Gemini users.

Who got it first, and why

The first group was a set of trusted cyber defenders, brought in through what Google calls its Fairwind Program. The plan was staged: security teams first, then paying developers and top-tier subscribers, then everyone else later.

The reason is the double edge. A model that is good at finding holes in software can help the people patching them, and it can just as easily help the people exploiting them. Giving defenders a head start is the responsible move.

What Google says it can do

Google's own examples are big. According to coverage of the announcement, its engineers used the model to untangle about 6 million lines of old Chrome code in a matter of weeks, and YouTube cut the processor load of its comments page in half. It's also described as handling an enormous amount of text at once, on the order of 1,500 pages.

Those are Google's numbers, from Google's own projects. Nobody outside the company could check them at launch.

How the story has moved since

Two weeks after the announcement, on Oct. 13, Google began rolling Argon out in the Gemini app to Google AI Pro and Ultra subscribers, with developers getting access through its API. Free users still don't have it. We cover who gets what, and what it costs, in the follow-up.

Independent testers have started to publish their own results. One roundup of public tests ranked Argon first on one list and eighth on another, a good reminder that how a model scores depends on what you measure.

What this means for you

If you use Gemini's free version, nothing has changed for you yet. If you pay for Google's AI plans, the follow-up has the details on what you can try. If you work in IT or security, the interesting question is how fast defenders can use tools like this compared with attackers.

My take: Two things are true here. Holding it back is the responsible move, because an AI that can find holes in software could help hackers just as easily as it helps defenders. But Google had also been playing catch-up with its rivals, and this announcement doubles as a "we're back" headline. So far the big performance claims come from Google, not independent testers. File this one under promising but unproven. If you use Gemini today, nothing changes for you yet.

Disclosure: Claude, made by one of Google's competitors, helped me research and write this.

Sources: The Stack, AI Career Lab, VKTR, AI Miracle, Let's Data Science

Hi, I'm Thomas

I've worked with my hands for 20+ years in industrial construction, electrical, automation and refrigeration. No programming background. I test AI tools after hours and tell you what actually works.

The site is free to read, and always will be.

Read my full bio →