Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use
Ars Technica Ryan Whitwam ● Covered by 6 sources
Google unveiled Gemini 4 Carbon, a newer AI model you still can’t use. It’s staged out slowly, but Google says it’s much better at code, security, and science.
Based on reporting by Ars Technica, Ryan Whitwam — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google is back with another frontier AI model, and this one arrives only two weeks after Gemini 4 Argon. The new model is called Gemini 4 Carbon, and it is not the Flash version people may have been waiting for. Instead, Google frames it as an upgrade to Argon, with a staged rollout that could still take some time.
The company is talking up Carbon’s performance in coding, agentic tasks, and scientific analysis. But there’s a catch: Google says the model is so capable that it cannot be opened up to users yet. That makes the announcement feel less like a product launch and more like a promise that comes with a delay.
Security is where Google is leaning hardest. The company says Carbon shows a significant leap in vulnerability detection, validation, and remediation. It also says the model is twice as likely to generate code without vulnerabilities as Argon.
And the Chrome team is already stacking on another claim: Carbon is said to be 10 times better at producing high-quality patches than non-Google models. Google adds that the model’s score on WebDev Arena is up by 89 points, which puts it at number four on the leaderboard. The pattern is familiar by now: a big reveal, a bigger set of claims, and a model that still isn’t in anyone’s hands.
My take — AI-written commentary, not fact-checked reporting
Google has turned frontier AI into a kind of trailer season: lots of dazzling claims, very little immediate access. The security pitch is the most interesting part, because “better at patching” is more useful than another vague intelligence flex. But if the model can’t be used yet, the hype machine is still doing most of the shipping.
Read more about this at: Ars Technica
