Google announces Gemini 4 Carbon with staged rollout

By Infinithink Studio · October 28, 2026

Google DeepMind announced Gemini 4 Carbon, which it calls its strongest model, with a focus on cybersecurity and long tasks. Access starts with a small group of cyber defenders.

Gemini 4 Carbon is the name of a new model announced on October 28, 2026 by Google DeepMind. According to Google, it is its strongest model so far, with a focus on cybersecurity and on long, multi-step tasks. Access is limited at launch.

What happened

According to Google DeepMind's announcement, Gemini 4 Carbon builds on Gemini 4 Argon, which Google says it launched two weeks earlier. Google describes Carbon as its strongest model, with gains in coding, agentic tasks and cybersecurity defense.

Google reports that Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59. For security work, Google cites three results on external benchmarks, evaluated without guardrails according to the company:

• 88.2% on SEC-bench Pro, which Google links to finding vulnerabilities.
• 54.8% on ExploitGym, linked to checking whether a vulnerability can be exploited.
• 71.0% on CWE-bench, linked to producing working code fixes.

Google also says that, on its internal benchmarks, Carbon more than doubles the success rate of Argon in generating secure code. These are figures declared by the company and we have not verified them independently.

On rollout, Google says it is taking a deliberate approach. It states that it is engaged in the voluntary pre-release testing process of the U.S. government, and that it is making Carbon available to a subset of cyber defenders. Access will then widen gradually to more trusted partners through its Fairwind Program. Google says it intends to open the model to developers, businesses and consumers through its product surfaces as soon as possible, without giving a date. Ars Technica, in its coverage, notes that Google makes no promise about the timeline for a wide release.

On pricing, Google states that Carbon will launch at the same introductory price as Argon, $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% less than the standard input price.

Why it matters

The announcement shows how quickly new frontier models are appearing. Google itself points to the two weeks between Argon and Carbon as an example of its pace.

The focus on cybersecurity is the distinctive part. Google says Carbon covers the defense cycle, from detecting a vulnerability to validating it and fixing it, and that it is reaching defenders first. Google also says it has worked on safety, citing a 0.0% attack success rate on an indirect prompt injection benchmark and a reduction of unsupported claims of about 2x compared with Argon in its internal tests.

Press coverage adds some context. Ars Technica observes that the gaps between frontier models on these benchmarks are a few points, and that benchmarks do not tell the whole story. The Verge reports that, for now, the model is available only to a select group of cybersecurity defense companies.

What it means for your business

For most small businesses, professionals and agencies, nothing changes this week. The model is not generally available, and the announcement does not give a date for wider access. This suggests that the practical effects will be visible later, once the product surfaces mentioned by Google are open.

It is reasonable to expect that tools built on top of newer models will gradually improve on long tasks and on code review. This is a hypothesis, not a fact reported by the sources. It is also reasonable to treat benchmark scores as a starting point. Results declared by a vendor depend on how the test was run, and they say little about your own documents, clients and processes.

If you work in a company that handles customer data, the cybersecurity angle is worth a conversation with whoever looks after your IT. Questions about which AI services are used, with which data and under which contract are valid whatever the model.

What to do now

• Note the news and wait for general availability. Check Google's official pages for updates before changing any tool or budget.
• Write down which tasks take your team the most time, such as summaries, replies to clients or first drafts. Test any new model on those tasks with your own material.
• If you use external AI services, check which data you send and what the provider's terms say, and involve a professional for legal or security questions.
• Keep a person in charge of reviewing what the AI produces before it reaches clients.

If you want to start with something practical, you can try the text summarizer on a long document of yours, or read what AI agents are to understand the tasks that models like this one are designed for. In both cases, the output is a draft that you review.

Sources

• Google DeepMind, Gemini 4 Carbon: Our next frontier model
• The Verge, Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity
• Ars Technica, Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use

For information only, this is not legal or professional advice. Information is current as of the publication date and may change. Official sources checked on October 28, 2026.