Google has a new AI model, and it isn’t aimed at everyday chat. On October 28, 2026, Google announced Gemini 4 Carbon, which it calls its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. In plain terms, it’s built for hard technical jobs like finding security flaws in software, checking whether they can be exploited, and writing the fixes.
Most people can’t use it yet. Google is starting with a small group of cyber defenders and will widen access over time. Here are five things worth knowing.
What Is Gemini 4 Carbon?
Gemini 4 Carbon comes from Google DeepMind and arrives about two weeks after Gemini 4 Argon. Google says it brings “significant gains” in coding and agentic tasks, as well as in cybersecurity defence.
“Agentic” just means the model can work through a long chain of connected steps with less hand-holding. Google says its own teams have watched Carbon work on tasks for days. It says that kind of stamina is what makes the model useful for big, messy engineering problems, not just quick answers.
Finding and Fixing Software Vulnerabilities
Security flaws in software are how attackers get at sensitive data and disrupt services. Finding them before the bad guys do is a huge part of defence, and it takes a lot of time and expertise. Google says Carbon covers the whole defence cycle:
- Vulnerability detection: spotting real weaknesses in software.
- Validation: checking whether a weakness can actually be exploited.
- Remediation: helping write working fixes.
Google also shared a couple of examples of the gains. In its internal tests, Carbon more than doubled Argon’s success rate at generating secure code. Google’s Chrome team reported that Carbon did more than 10 times better than another lab’s large frontier model at producing high-quality patches.
One caution applies, as with any AI-generated security work. Findings and code fixes still need proper testing and human review before they go anywhere near critical systems.
Benchmark Scores: What Google Reports
Here are the numbers Google has shared:
| Benchmark | Reported Score | What It Measures |
| Artificial Analysis Intelligence Index v5.0 | 59 (ranked #1) | Overall AI performance |
| SEC-bench Pro | 88.2% | Finding vulnerabilities |
| ExploitGym | 54.8% | Exploiting vulnerabilities |
| CWE-bench | 71.0% | Patching vulnerabilities |
Take these with a pinch of salt. They are Google’s own reported results, and Google says the external cybersecurity tests were run “without guardrails”. That means customers won’t necessarily see the same scores once safeguards are in place. A benchmark also isn’t the same as a real-world workday.
Google adds that Carbon makes about half as many unsupported claims as Argon in its internal testing, and that it shows lower misalignment and fabrication rates in live coding work.
Why Can’t Everyone Use It Yet?
Because of how powerful it is. Google is starting with a subset of cyber defenders and will gradually add trusted partners through its Fairwind Program. SecurityWeek reports that more than 650 organisations currently take part in Fairwind.
Google says it’s still gathering feedback and refining its guardrails before opening the model up further. That fits the general worry with security AI: a tool that helps defenders find flaws could help attackers too. Google also says it’s engaging with the U.S. government’s voluntary process for pre-release model testing. Google says Carbon is coming soon to developers, businesses and consumers, but it hasn’t given a date. So for now, most of us will have to wait.
Pricing
Google has announced introductory API pricing that matches Argon’s:
| Pricing Details | Cost |
| Input tokens | $2 per million |
| Output tokens | $10 per million |
| Cached input tokens | 95% off the input price |
After the introductory period, the prices go up to $4 per million input tokens and $20 per million output tokens. Google hasn’t published when that period ends. These are developer API prices, not subscription prices for the regular Gemini app.
Final words
Carbon shows where AI is heading. It’s moving beyond chat and text towards models that do long, complicated technical work, including protecting software. For businesses, that could mean faster vulnerability detection and maintenance. But the cautious, phased rollout is a reminder that a model this capable needs careful handling, and that humans still need to stay in the loop. How good Carbon really is will become clearer once more security teams and developers get to try it.
FAQs
- What is Google Gemini 4 Carbon?
Gemini 4 Carbon is Google’s latest advanced AI model, designed for complex coding, software engineering and cybersecurity tasks.
- What are the main features of Gemini 4 Carbon?
Its key features include software vulnerability detection, security risk assessment, code generation, vulnerability fixing and long-running engineering tasks.
- Is Gemini 4 Carbon available to everyone?
No. Google is initially providing access to selected cybersecurity professionals and trusted partners through its Fairwind Program.
- What is the price of Gemini 4 Carbon?
Its introductory API pricing is $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95% discount.
- How is Gemini 4 Carbon different from Gemini 4 Argon?
Carbon builds on Argon with improvements in coding, cybersecurity, agentic capabilities and secure-code generation, according to Google.
- Can Gemini 4 Carbon detect and fix security vulnerabilities?
Yes. Google says the model can identify software weaknesses, assess their potential exploitability and help developers develop code fixes.
