Independent Minds. Critical Thinking. A Better World.

Science and Technology

Gemini 4 Carbon: Google gives its cyber-focused AI to defenders first

Gemini 4 Carbon is Google’s newest frontier model, and it will not reach the public yet. Google announced it on 28 October.

The company calls it its strongest model, with leading cybersecurity and long-horizon skills. It is rolling the model out slowly, starting with a small group of cyber defenders.

How Gemini 4 Carbon will be released

Access will widen to more trusted partners through what Google calls the Fairwind Program. Developers, businesses and consumers come later. Google says it first wants to refine its guardrails.

The rollout is part of a voluntary US government pre-release testing process. That detail matters. Few labs have openly tied a launch to outside testing.

What Google says it can do

Google published benchmark scores that it ran without guardrails. It reports 88.2 percent on SEC-bench Pro, which tests vulnerability finding. It reports 54.8 percent on ExploitGym, which tests exploitation. It reports 71.0 percent on CWE-bench, which tests patching.

The company also says the model writes secure code at more than twice the success rate of its predecessor, Gemini 4 Argon. These are Google’s own figures. Independent testing has yet to confirm them.

AI Weekly adds that the model scores 59 on the Artificial Analysis Intelligence Index, the top spot on that list.

Price and the dual-use problem

The introductory price is $2 per million input tokens and $10 per million output tokens. Google says it will rise to $4 and $20 after the introductory period.

A model that finds and exploits flaws can help defenders. It can also help attackers. That is the dual-use problem, and it explains the staged release.

Regulators are watching AI labs closely. See our report on the FTC demand for Anthropic executive testimony.

Expect other labs to follow with gated releases. The test will be whether outside experts can check the claims before the model goes wide.

Fact basis: Google announcement (28 October 2026) and AI Weekly summary (28–29 October 2026). Benchmark figures are Google’s own and unverified.

0 0 votes
Article Rating
Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x