Google says its new Gemini 4 Carbon model is designed for complex coding and cybersecurity work, but access is initially limited to selected cyber defenders as the company continues testing its safeguards.
Google has unveiled Gemini 4 Carbon, a new artificial intelligence model it describes as its strongest model yet for cybersecurity and long-running engineering tasks. The launch adds to the company’s Gemini 4 family and intensifies competition among major AI developers working on advanced coding systems.
Announced on 28 October, Carbon is designed to work through complicated, multi-step engineering problems, including finding software vulnerabilities and helping developers patch them. Google says the model also improves on its predecessor, Gemini 4 Argon, in several coding and cybersecurity evaluations.
What can Gemini 4 Carbon do?
According to Google, Carbon is built for tasks that require sustained work rather than a single response. These include tracing difficult bugs across large codebases, identifying weaknesses in software and helping produce fixes.
Google reported scores of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for vulnerability exploitation evaluation, and 71.0% on CWE-bench for patching weaknesses. These are benchmark results reported by Google; they should not be treated as a guarantee that the model will perform equally well in every real-world environment.
The company also said Carbon achieved more than twice the success rate of Gemini 4 Argon at generating secure code on its internal benchmarks. Those figures are company-reported results and may depend on the test design and evaluation conditions.
Why is the model focused on cybersecurity?
AI systems can help security teams examine large amounts of code, identify possible vulnerabilities and prioritise fixes. Used responsibly, such capabilities could help defenders find and repair weaknesses before attackers exploit them.
The same capabilities can also create risks if misused. Tools that identify vulnerabilities or assist with exploitation need safeguards, access controls and responsible deployment. Google says Carbon is being introduced gradually while it gathers feedback and continues refining its protections.
Can everyone use Gemini 4 Carbon now?
No. Google says initial access is being provided to a subset of cybersecurity defenders, with additional trusted partners expected to join through its Fairwind Program. The company says it plans to make the model available to developers, businesses and consumers through its products after further evaluation and iteration.
That means everyday users should not assume Carbon is already available in the regular Gemini app. Availability may change as Google expands access.
How much will it cost?
Google announced introductory API pricing of US$2 per million input tokens and US$10 per million output tokens, matching the introductory pricing for Gemini 4 Argon. The company says those rates will later increase to US$4 per million input tokens and US$20 per million output tokens after the introductory period.
These are API token prices, not a monthly consumer subscription. The actual cost of using a model depends on the amount of input and output processed and the service through which it is accessed.
What does this mean for developers and businesses?
For software teams, the main promise is help with complicated tasks that can take considerable time to diagnose and resolve. For cybersecurity teams, the potential value lies in speeding up vulnerability discovery, validation and remediation—provided that human experts verify the findings and control any changes made to production systems.
Businesses should evaluate such systems against their own workflows, security requirements, costs and reliability standards rather than relying on headline benchmark scores alone. AI can assist security professionals, but it does not remove the need for experienced developers, security reviews or incident-response plans.
The bigger picture
Gemini 4 Carbon reflects the growing overlap between advanced AI, software engineering and cybersecurity. As these models become more capable, the challenge for developers is not only to improve performance but also to ensure that powerful capabilities are deployed responsibly.
For now, Carbon’s rollout is deliberately limited. Its broader impact will depend on how it performs outside benchmark tests, how access expands and whether the model can help defenders improve security without introducing unacceptable new risks.
Sources
- Google DeepMind: Gemini 4 Carbon announcement
- Reuters: Google launches Gemini 4 Carbon with cybersecurity focus
GAWAH Technology | GAWAH News Desk
