Breaking

Google Introduces Gemini 4 Carbon With a Cybersecurity Focus

October 28, 2026 at 13:41 EDT

Foundation Models
Security
AI Agents

Google announced Gemini 4 Carbon on Wednesday as its newest frontier model, emphasizing cybersecurity and sustained, long-horizon work. It follows Gemini 4 Argon, introduced in September, and is designed for engineering tasks that can take multiple steps or run for days. Gemini 4 Argon

OCTOBER 28, 2026 · GOOGLE

Gemini 4 Carbon targets cybersecurity

Google’s new frontier model pairs long-running engineering work with a focus on finding and fixing software vulnerabilities.

88.2%

Vulnerability finding
SEC-bench Pro

54.8%

Exploit validation
ExploitGym

71.0%

Vulnerability patching
CWE-bench

Token rates double after the introductory period

Each pair uses its own scale. Columns show the listed price per million tokens.

Input tokens

Intro

Later

Output tokens

$10

Intro

$20

Later

SAFETY AND CONTROL

Capability gains come with guardrail evaluations

Google reports fewer coding shortcuts and stronger resistance to prompt injection in its evaluations.

0.0%Successful cheating on Terminal-Bench 4.0

0.0%Attack success on Gray Swan’s indirect prompt-injection benchmark

What early testers liked

They described stronger 3D output than Argon and more persistence on complex tasks.

Where they saw friction

Reports flagged slower runs and weaker frontend and interface work; comparisons used selected prompts rather than standardized tests.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.