Google Launches Gemini 4 Carbon with Cybersecurity Focus and Restricted Access
Google DeepMind unveiled Gemini 4 Carbon on October 28, 2026, a frontier AI model that prioritizes cybersecurity defense and long-horizon coding tasks. Arriving just two weeks after Gemini 4 Argon, Carbon ranks first on the Artificial Analysis Intelligence Index but remains restricted to a subset of cyber defenders rather than general enterprise users.
How does Gemini 4 Carbon perform on cybersecurity benchmarks?
The model achieves state-of-the-art performance across the cybersecurity defense lifecycle. It scores 88.2% on SEC-bench Pro for autonomously uncovering complex flaws in real-world software. For validating risk, it reaches 54.8% on ExploitGym, and for patching root causes, it hits 71.0% on CWE-bench.
Google notes that Carbon generates secure code at more than twice the success rate of its predecessor on internal benchmarks. Additionally, the Chrome team reported that Carbon is 10 times better at generating high-quality patches compared to a competing lab’s model, demonstrating a significant leap in automated remediation capabilities.
What are the pricing and context window for Gemini 4 Carbon?
The model features a 1 million token context window, allowing it to process massive codebases and extensive documentation in a single prompt.
Google is offering an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at a 95% discount. However, Artificial Analysis notes that the standard pricing will be $4.00 per million input tokens and $20.00 per million output tokens once the launch discount ends. IT leaders budgeting for AI integration should account for this standard tier when projecting long-term operational costs.
Why is Google restricting access to the Fairwind Program?
Google is taking a deliberate approach to this rollout, actively engaging in the U.S. government’s voluntary process for pre-release model testing. Access is currently limited to a subset of cyber defenders, expanding gradually through the Fairwind Program.
Koray Kavukcuoglu, SVP at Google DeepMind, described the release as the:
"strongest model with state-of-the-art cybersecurity and long horizon capabilities"
The rapid two-week release cycle follows Google’s strategy where researchers and models are "working hand in hand to advance capabilities faster", while simultaneously iterating on safety guardrails. Reuters reported this launch steps up competition with OpenAI and Anthropic as the race to build more capable coding and cybersecurity systems intensifies.
What does Gemini 4 Carbon mean for enterprise security teams?
For IT decision-makers, the practical implication is a fundamental shift in how vulnerability management can be automated. A model that can independently find, validate, and patch vulnerabilities moves AI from a passive code-completion tool to an autonomous security agent.
However, because Carbon is not yet available via standard API endpoints, enterprise security teams cannot immediately integrate it into their existing CI/CD pipelines. Organizations must either wait for general availability or apply for the Fairwind Program to test these agentic security workflows in controlled environments. The model’s ability to “handle complex engineering tasks that can take days to complete” indicates that its full-stack web development capabilities are maturing rapidly for sustained workflows. Furthermore, internal testing shows Carbon achieves reduced misalignment and fabrication rates, providing a measurable baseline for enterprises evaluating the reliability of autonomous code patching.
Related articles
