Google Gemini 4 Carbon Begins Phased Rollout for Long-Horizon AI Tasks

Google has introduced Gemini 4 Carbon, a new Gemini 4 frontier model positioned for sustained, long-horizon work such as complex software tasks, coding and cybersecurity. The important qualification for businesses is availability: Carbon is entering a phased pre-release program, not a single-day global launch. Google says it will initially test the model with the U.S. government and trusted cyber defenders, then expand access gradually to more partners before broader availability for developers, businesses and consumers.

In Google's official Gemini 4 Carbon announcement, the company describes Carbon as its strongest Gemini 4 model and a successor that builds on Gemini 4 Argon. Google also says Carbon will launch at Argon's introductory pricing: $2 per 1 million input tokens and $10 per 1 million output tokens. Cached input tokens are priced at a 95% discount from the input price.

For teams considering advanced AI work, Carbon's announcement is most relevant as a signal of where the Gemini roadmap is heading. Its stated focus is not simply producing a better one-off answer. It is maintaining reasoning across longer, multi-step tasks where earlier outputs and decisions need to inform subsequent work.

What Gemini 4 Carbon changes

Google presents Carbon as a model for difficult, persistent tasks that require deeper reasoning over a longer sequence of actions. That orientation is particularly visible in its cybersecurity testing, where the company reports results across vulnerability discovery, exploitation and patching.

Google reports the following benchmark results for Carbon:

• 88.2% on SEC-bench Pro for finding vulnerabilities.
• 54.8% on ExploitGym for exploitation.
• 71.0% on CWE-bench for patching.

These are company-reported benchmark figures, rather than evidence of outcomes in every real-world environment. Still, they illustrate the kinds of workflows Google is using to evaluate the model: technical tasks that involve investigation, multiple decisions and follow-through, rather than a single prompt and response.

Google also cites improvements over Argon in areas including coding, knowledge work and WebDev Arena. The company emphasizes continued safety guardrails, along with safer and more transparent outputs from its internal testing. Carbon's initial deployment through government and trusted-defender testing reflects the sensitivity of advanced cybersecurity capabilities and the importance Google places on controlled evaluation before wider access.

Availability is the immediate constraint

The practical near-term question is not whether Carbon has been announced, but who can use it. Google says initial testing begins with the U.S. government and a cohort of trusted cyber defenders through its Fairwind Program. It plans to expand to additional trusted partners before opening access more broadly.

That means organizations outside the early cohorts should not assume immediate production access. Google has stated the direction of travel, but the supplied announcement does not provide a date for broad developer, business or consumer availability. Teams that currently use Gemini should therefore treat Carbon as an upcoming option and continue planning around tools they can access today.

The phased approach also makes early feedback consequential. Google can evaluate how the model performs in demanding settings before extending availability, especially where sustained reasoning and cybersecurity work require careful safeguards.

Pricing and practical planning

Carbon's announced introductory rates create continuity with Argon: $2 per million input tokens and $10 per million output tokens. A 95% discount on cached input tokens may be relevant for workflows that repeatedly reuse the same material, such as recurring context or instructions.

Token pricing alone does not determine a project's total cost. The volume of prompts, length of inputs and outputs, and the number of steps in a workflow all affect usage. For businesses, the useful preparation now is to identify work where long-running reasoning could matter most, then define a measurable outcome before access expands.

Potential candidates include:

• Investigating and prioritizing software issues across a complex codebase.
• Supporting multi-step technical analysis that needs to retain prior context.
• Reviewing recurring knowledge-work processes where teams assemble information, make decisions and produce an output over several stages.
• Assessing security workflows where human experts remain responsible for validation and action.

Carbon's benchmark emphasis should not be read as a reason to hand over security decisions without review. The announcement describes testing and safeguards, while real business use will still depend on the specific workflow, data and human checks involved.

For companies planning AI adoption, the broader lesson is to design around a process, not a model name. Map the steps that consume time, decide what information the system can use, establish who validates its output, and measure whether the workflow improves. That preparation can make a new model easier to evaluate when it becomes available.

If long-running AI tasks could reduce manual analysis or accelerate technical workflows in your business, Scalevise can help turn that possibility into a focused implementation plan. Our practical AI adoption consultancy helps teams identify viable use cases, assess the tools they can use today and build a roadmap around measurable operational outcomes rather than model hype. Request an AI consultancy conversation to prioritize your next workflow.

Frequently Asked Questions

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's newly announced Gemini 4 model, designed for sustained, long-horizon reasoning across complex tasks such as coding and cybersecurity.

Is Gemini 4 Carbon broadly available now?

No. Google describes a phased rollout that starts with pre-release testing involving the U.S. government and trusted cyber defenders, followed by gradual expansion to more partners and later broader availability.

How much will Gemini 4 Carbon cost?

Google says Carbon will use the same introductory price as Argon: $2 per 1 million input tokens and $10 per 1 million output tokens. Cached input tokens receive a 95% discount from the input price.

What cybersecurity results has Google reported for Carbon?

Google reports 88.2% on SEC-bench Pro for vulnerability finding, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching.

Conclusion

Gemini 4 Carbon is a confirmed addition to Google's Gemini 4 roadmap, with a clear focus on complex work that unfolds across multiple steps. Its phased release means most organizations should watch for broader access rather than expect immediate use, while its Argon-level introductory pricing provides an early reference point for future evaluation. The most useful preparation is to identify the workflows where sustained reasoning could create a measurable benefit.