Google has introduced Gemini 4 Carbon, a new AI model focused on cybersecurity and advanced coding. Announced this October 2026, the model can identify software vulnerabilities, test potential exploits, and develop code fixes.
Google is initially limiting access to selected cybersecurity defenders and trusted partners. The launch follows Gemini 4 Argon, introduced just two weeks earlier.
What Is Gemini 4 Carbon?
Gemini 4 Carbon is Google’s latest frontier AI model, built to handle demanding engineering tasks that require extended reasoning and multiple steps. It builds on Gemini 4 Argon, with improvements in coding, agent-based workflows, and cybersecurity.
According to Google, Carbon can work on complex engineering challenges for days, including identifying the root causes of difficult bugs and developing fixes across large codebases and low-level systems.
The model also ranked first on the Artificial Analysis Intelligence Index v5.0, scoring 59. This adds to Google’s reported performance improvements across coding, scientific tasks, and software development.
Reuters reported that the launch intensifies competition among Google, OpenAI, and Anthropic in advanced AI coding and cybersecurity systems.
How Does the AI Model Identify Security Vulnerabilities?
AI vulnerability detection involves identifying software weaknesses that attackers could exploit. Gemini 4 Carbon is designed to support three important stages of the cybersecurity defence process:
- Vulnerability detection: Identifies potential security weaknesses in real-world software code.
- Exploit validation: Tests whether identified vulnerabilities can be exploited, helping security teams assess their actual risk.
- Code remediation: Develops code fixes designed to address vulnerabilities at their root cause.
Google also reports that Carbon achieved more than twice Argon’s success rate in generating secure code on internal benchmarks. This improvement applies to Google’s specific evaluation and does not mean the model is twice as capable across every coding task.
Gemini 4 Carbon Benchmark Results
Google reported the following results across three cybersecurity benchmarks, evaluated without guardrails:
| Benchmark | Reported score | Purpose |
| SEC-bench Pro | 88.2% | Finding software vulnerabilities |
| ExploitGym | 54.8% | Evaluating exploitability |
| CWE-bench | 71.0% | Developing vulnerability patches |
The company also reported a 0% attack success rate on Gray Swan’s Indirect Prompt Injection (IPI) benchmark, compared with 0.7% for Gemini 4 Argon.
These results represent performance on specific evaluations. They do not establish that Carbon can identify every software vulnerability, prevent all cyberattacks, or resist every form of malicious instruction.
Who Can Access Gemini 4 Carbon?
Google is taking a phased approach to the model’s release. Initial access began with a selected group of cybersecurity defenders, with additional trusted partners expected to receive access through its Fairwind Program.
The programme allows Google to gather feedback, evaluate potential misuse, and improve safeguards before expanding availability. The company is also participating in the US government’s voluntary process for pre-release AI model testing.
This restricted rollout reflects the dual-use nature of advanced cybersecurity AI. The same capabilities that help security professionals identify and fix software weaknesses could also be misused by malicious actors.
When Will Gemini 4 Carbon Be Available to Everyone?
Google plans to make Gemini 4 Carbon available to developers, businesses, and consumers through its products. However, it has not announced a confirmed date for general availability.
The company has announced introductory API pricing of $2 per million input tokens and $10 per million output tokens, matching Gemini 4 Argon’s introductory rates.
Cached input tokens receive a 95% discount on the input token price. After the introductory period, standard rates will increase to $4 per million input tokens and $20 per million output tokens.
These are usage-based API charges rather than monthly subscription prices.
What Does This Mean for Cybersecurity?
Gemini 4 Carbon could help security teams investigate software weaknesses, prioritise vulnerabilities and accelerate code remediation. Its reported benchmark performance also highlights the growing role of AI in software security.
However, benchmark results alone cannot establish how reliably a model will perform in every production environment. Human review, independent testing, and established security controls will remain important.
For now, Gemini 4 Carbon’s significance lies in its reported cybersecurity capabilities and controlled rollout. Its broader impact will depend on real-world performance, safety measures, and wider access as Google expands its AI offerings.
