Google launched Gemini 4 Carbon: the model that hunts vulnerabilities at $2 per million tokens
Direct answer
On October 28, 2026, Google introduced Gemini 4 Carbon, its new frontier model, with a stated focus on cybersecurity, coding and complex reasoning, per Reuters. It arrives two weeks after Gemini 4 Argon. Its price is $2 per million input tokens and $10 per million output tokens, matching Argon's launch price. First availability is limited: Google offers it first to a subset of cybersecurity defenders in its Fairwind Program while it continues testing safeguards.
What makes Carbon different
- Security as a core task: Google says Carbon delivers state-of-the-art performance on cybersecurity tasks including vulnerability discovery, exploitation and software patching.
- Safer code, on internal benchmarks: Google says Carbon generates secure code at more than twice Argon's success rate on its internal benchmarks. That figure comes from the company itself, and it is best read that way until comparable independent evaluations exist.
- Long engineering tasks: the model is built for complex, long-running work of the kind that takes hours or days, not only fast answers in a chat window.
- The same price as Argon: $2/$10 per million tokens matches Argon's launch price. Frontier competition is also fought on price, and Google holds the same reference for two launches in a row.
Limited access says as much as the announcement
- Defenders first: giving Carbon to cybersecurity defenders first makes sense: a model strong at finding vulnerabilities helps defenders, and it could also help attackers. Google says it is still testing safeguards before widening access.
- The question for your company: if a vendor offers you an AI security tool, ask what it can do in your code, what logs it leaves, who reviews its findings, and what happens when a false positive blocks a deployment.
- A finding still needs a human: a vulnerability reported by a model is a hypothesis with evidence, not a verdict. Patching, testing and the decision to deploy remain your team's work.
- What to watch: access beyond Fairwind, pricing after the launch period, and independent evaluations that confirm or qualify Google's internal figures.
What it means for your business
- Security as an internal capability: if you already use AI to review code, Carbon reinforces the trend: security review becomes a capability of the model, not only a separate tool run at the end.
- Predictable budget: at $2/$10 per million tokens, experimenting is cheap compared with a frontier model from a year ago. The real cost sits in human review and integration, not only in tokens.
- Before adopting it: run a trial on your own repository, with your own known vulnerabilities, and measure how many real findings it delivers versus false positives. That trial is worth more than any vendor benchmark.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- Google launched Gemini 4 Argon at launch pricing
- OpenAI launched GPT-6.1 Terra at half price
- Claude Haiku 5.5 costs up to 90% less
Frequently asked questions
What is Gemini 4 Carbon?
It is Google's frontier model introduced on October 28, 2026, focused on cybersecurity, coding and long engineering tasks. It arrives two weeks after Gemini 4 Argon.
How much does Gemini 4 Carbon cost?
$2 per million input tokens and $10 per million output tokens, matching Gemini 4 Argon's launch price.
Can I use Gemini 4 Carbon already?
Not generally yet. Google offers it first to a subset of cybersecurity defenders in its Fairwind Program while it continues testing safeguards.
Does a model like this replace my security team?
No. It finds and proposes, but patching, testing and the deployment decision remain your team's responsibility. Model findings need human verification.
