Google's Gemini 4 Carbon goes to cyber defenders first

The best model Google has made is one you can't use yet.

Google announced Gemini 4 Carbon on Oct. 28. In the post, Koray Kavukcuoglu, SVP at Google DeepMind and Google's chief AI architect, calls it Google's "strongest model with state-of-the-art cybersecurity and long horizon capabilities." It arrives two weeks after Gemini 4 Argon. Access starts this week for a subset of cyber defenders.

Defenders get it first, everyone else later

Google says access will widen gradually to more trusted partners through its Fairwind Program. Developers, businesses and consumers get it "as soon as possible," after more feedback and more work on guardrails. Google gave no date.

Google also says it is taking part in the U.S. government's voluntary pre-release model testing process.

The price is already set. Intro API pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the intro period, it doubles.

$4 and $20.

Google didn't say when the intro period ends.

The benchmarks are about breaking and fixing code

Google's numbers lean on security, and they are Google's numbers. It reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them and 71.0% on CWE-bench for patching them.

On general ability, Google says Carbon scores 59 on the Artificial Analysis Intelligence Index v5.0, ranked first. When Artificial Analysis published v5 on Oct. 27, Claude Opus 5.5 led at 58 and Gemini 4 Argon scored 50. If Google's figure holds, Carbon is a nine-point jump over its own sibling in two weeks.

Google also reports behavior changes from its internal testing. It says unsupported claims fell by roughly half compared with Argon, and high-severity misalignment alerts in live coding environments dropped by almost 80%. On Gray Swan's indirect prompt injection benchmark, it reports a 0.0% attack success rate, against 0.7% for Argon. On Terminal-Bench 4.0, it says successful cheating fell to 0.0%.

Why the gate matters more than the score

Here's the thing: a model that is best at finding and exploiting bugs helps whoever holds it. Google chose to put it in defenders' hands before anyone can rent it by the token.

That choice has a cost for you. The model at the top of the index has no date for the Gemini API. Argon, which TopFive covered when it went half price until Nov. 30, is still the Gemini 4 model you can build on today.

It also fits the month. AI-found bugs have been forcing patch cycles, including PHP's fixes for AI-found flaws. A lab giving its best bug finder to the people doing the patching is betting that defense needs a head start.

Google's CEO is already counting it. In Alphabet's third-quarter release, Sundar Pichai said: "Gemini 4 Argon and Carbon are now leading at the frontier."

Watch two dates

The first is when Carbon reaches the Gemini API. The second is when the intro price ends.

Plan at $4 and $20. Anything cheaper is a bonus.