Gemini 4 Carbon lands weeks after Argon, and almost nobody can use it

Google announced Gemini 4 Carbon on October 28 and calls it its strongest model yet. The timing stands out: it arrives about two weeks after Gemini 4 Argon.

The focus is cybersecurity and long running tasks, the kind where a model has to work for a long time without losing track.

Carbon is a specialized Argon, not a successor

By Google's own numbers, Carbon tops the Artificial Analysis Intelligence Index with a score of 59 on version 5.0.

On security, the company cites 88.2% on SEC-bench Pro, a test of finding vulnerabilities in code. It also claims an edge in exploiting and patching flaws.

These are vendor results. No independent lab has reproduced them yet.

The gap with Argon is less about a general leap and more about focus. Argon is the broad model. Carbon targets defense teams that need to comb through large systems looking for holes.

For developers, the reported introductory price matches Argon at $2 per million input tokens. Few people will get to use it at that price anytime soon.

That explains the rollout. Access starts with cybersecurity professionals and then widens through a Google program called Fairwind.

There is no date for the general public. As of the 29th, it was also missing from the Gemini API model list, according to an independent check.

Google says it takes part in the US government's voluntary pre-release testing process. That suggests the company treats the model as sensitive, since the same skill that finds bugs to fix can find bugs to exploit.

My bet is that Carbon works as a showcase more than a product, at least for now. People using Gemini on their phone or in the browser will not notice a difference anytime soon.

What the launch does show is the new race between Google, OpenAI and Anthropic: shipping specialized versions weeks apart to fight for the top of the leaderboards. The test that matters comes when outsiders can use it and measure it.