Google's New AI Is So Powerful It Won't Be Released Yet

Google has a new AI model called Gemini 4 Carbon, and it's a big deal for anyone who uses software—which is all of us. Carbon is Google's latest attempt to build the most powerful AI, and it's especially good at two things: writing computer code and finding security holes. Google says Carbon is twice as likely to write code without vulnerabilities compared to its previous model, Argon. That means the apps and websites you use could become safer and more reliable.

But here's the catch: Google isn't letting everyone use it just yet. Because Carbon is so good at finding and fixing security flaws, Google is taking a careful, staged approach. It's working with the US government's voluntary testing framework and will first roll it out through its Fairwinds cyber defense program. This is similar to how Google handled Argon, which took a couple of weeks to reach a wider audience. So don't expect to try Carbon tomorrow—but it's coming.

When it does arrive, Carbon will likely be available first to AI Ultra subscribers and developers using Google's API. And there's a money angle: Google is offering a temporary discount for developers—$2 per million input tokens and $10 per million output tokens—but that price will double after the introductory period. For context, that's like getting a half-price sale on a tool that helps build software, but you need to act fast because the sale won't last forever.

In tests, Carbon outperformed competing models from OpenAI and Anthropic in almost every category, including coding and security. It also resisted all attempted prompt-injection attacks—a type of hacking where someone tries to trick the AI into misbehaving. That's a huge improvement over other models, which had small but nonzero failure rates. So while Carbon isn't available yet, it's a sign that AI is getting better at protecting us from digital threats—and that's something we can all look forward to.