Access begins with a limited cohort
Google has announced Gemini 4 Carbon, a frontier model focused on cybersecurity and long-running engineering work, in its model announcement. Access starts with a subset of cyber defenders, followed by additional trusted partners through the Fairwind Program.
Availability for developers, businesses and consumers across Google’s product surfaces is still forthcoming. This is therefore a model announcement with a restricted rollout, rather than evidence that every Gemini customer can select Carbon today.
Reported gains have specific boundaries
Google reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench. The announcement states that these external cybersecurity benchmark results were evaluated without guardrails. They measure different tasks, so they should not be merged into a single percentage representing production security.
The company also reports improvements in secure-code generation, internal coding behavior and resistance to prompt injection. A zero attack-success result on one evaluation does not establish immunity to attacks outside that evaluation. The useful reading is a set of reported test outcomes with defined conditions, not a general security guarantee.
Price and deployment are separate questions
Google says Carbon will launch at the same introductory token prices as Argon: $2 per million input tokens and $10 per million output tokens. Cached input is advertised at a 95% discount to the input rate. These are announced introductory terms, while wider availability remains pending.
For teams planning a deployment, the next step is to establish whether they can access the model and which safeguards apply to that access. An evaluation should then use the team’s own tasks, tool permissions and review criteria. Google’s report gives reasons to investigate Carbon, but it does not replace a matched test of a real engineering workflow.