Google Unveils Gemini 4 Carbon: Its Strongest Cybersecurity Frontier Model, Defenders First

What happened

Google positions Carbon in its official blog as the next frontier model in the Gemini 4 family: the company calls it its strongest model yet, built for cybersecurity and long-horizon capability, following Argon, launched two weeks earlier, with internal teams already running it for days on debugging and patching tasks across large codebases and low-level systems.

Distribution starts through a restricted channel: a subset of cyber defenders gets access starting this week, followed by additional trusted partners through the Fairwind Program; Google says it is engaged in the US government's voluntary pre-release model testing process and will open Carbon to developers, businesses and consumers as soon as possible after gathering feedback and iterating on guardrails, with no date set for general availability.

Pricing carries over Argon's introductory rate: $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off; a footnote in the announcement states that after the introductory period ends, pricing rises to $4 per million input tokens and $20 per million output tokens, with the length of the introductory period undisclosed.

Key facts

1. Announced by: Google announced Gemini 4 Carbon on its official blog on Oct 28, reported by Reuters the same day.
2. Outside ranking: Carbon ranks first on Artificial Analysis' Intelligence Index v5.0 with a score of 59; Google calls it its strongest model yet.
3. Cybersecurity benchmarks (Google figures, evaluated without guardrails): 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation validation, 71.0% on CWE-bench for patching.
4. Internal figures: success rate for generating secure code more than doubles over Argon; Google's Chrome team says its high-quality patch generation beats another lab's large frontier model by more than 10 times.
5. Safety figures (Google internal testing): cheating success on Terminal-Bench 4.0 falls to 0.0%, indirect prompt injection attack success on Gray Swan's benchmark is 0.0%, and high-severity misalignment alerts drop by almost 80% versus Argon.
6. First channel and next steps: a subset of cyber defenders gets access this week, expanding through the Fairwind Program to trusted partners; no date is set for general availability to developers, businesses and consumers.
7. Pricing: introductory $2 per million input tokens and $10 per million output tokens, cached input at 95% off; rising to $4 and $20 after the introductory period.