TL;DR

Google announced Gemini 4 Carbon on 28 October, two weeks after Gemini 4 Argon, and describes it as its strongest model, with its biggest gains in cybersecurity and long-running engineering work. The first users will be a selected group of cyber defenders. Developers, businesses and consumers will get access later, once Google has gathered feedback and refined its guardrails.

What Google claims

The figures are Google’s own and were measured without guardrails. Carbon scores 59 on version 5.0 of the Artificial Analysis Intelligence Index, which Google says puts it first. On security benchmarks 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.

Internally, Google says Carbon more than doubles Argon’s success rate at writing secure code, and that its Chrome team found it more than ten times better at producing high-quality patches than an unnamed rival lab’s large frontier model. Teams have also run it for days at a time on multi-step jobs such as tracing and fixing subtle bugs in large codebases.

Google also reports alignment changes. Successful cheating on Terminal-Bench 4.0 fell to zero, high-severity misalignment alerts in live coding dropped by almost 80%, and unsupported claims were roughly halved compared with Argon. On Gray Swan’s prompt injection benchmark, every evaluated attack failed.

Pricing matches Argon’s introductory rate: $2 for each million tokens sent in and $10 for each million generated.

Why the staged release

Google says it is taking part in the US government’s voluntary pre-release testing process, and will extend access through its Fairwind Program to further trusted partners in stages. In our view, the logic is that a model this good at finding and exploiting flaws is as useful to attackers as to defenders, so the defenders get a head start.

We covered Argon’s launch and its general availability. Google presents the two-week gap between the two models as a sign of the pace at which it is now moving.

Looking Forward

For UK businesses, Carbon is not yet something to buy, and no date has been given for wider access. In our view, the release pattern matters more than the model. If other labs follow Google in routing their most capable systems through security partners first, organisations without established security relationships may wait longest for the newest tools, while the benchmark gains on patching suggest that defenders who do get access could shorten remediation times considerably.