Google is expanding access to Gemini 4 Carbon, its latest advanced AI model designed for complex reasoning, software engineering, scientific research, and cybersecurity. The model is now available to developers, enterprise customers, researchers, and everyday users through several Google products, following its earlier introduction alongside Gemini 4 Argon.

Gemini 4 Carbon is built to handle demanding tasks that require more than generating text or answering straightforward questions. Google says the model brings improvements in agentic coding, scientific and mathematical reasoning, and cybersecurity analysis, making it particularly useful for workflows involving complex problems and large amounts of information.

Gemini 4 Carbon focuses on advanced coding and cybersecurity

One of the biggest areas of improvement is software engineering. Google says Carbon can work through large codebases, identify complex bugs, and help developers investigate security vulnerabilities.

Google has also highlighted its cybersecurity capabilities, including the ability to analyze interconnected vulnerabilities and determine whether potential weaknesses can actually be exploited. The company says Carbon performed strongly in internal evaluations and external testing with cybersecurity partners.

Several security companies have shared positive early experiences with the model. Cisco, for example, highlighted Carbon’s ability to examine complex codebases and distinguish genuinely exploitable vulnerabilities from less meaningful findings. Other organizations, including Armadin, Chainguard, and Fortinet, have also reported promising results from their early evaluations.

However, Google is emphasizing defensive cybersecurity applications, with additional safeguards designed to prevent the model’s advanced capabilities from being misused.

Stronger scientific and mathematical reasoning

Beyond software development, Gemini 4 Carbon is designed to help researchers work through complicated scientific problems. Google says the model has improved reasoning capabilities across physics, mathematics, and life sciences, including analyzing experimental data, testing hypotheses, and handling multistep research tasks.

Google points to its work with Axiom Bio, a drug toxicity research company, as one example. According to the company, Carbon identified previously undiscovered mechanisms of toxicity, helping researchers investigate potential risks during drug development.

Google’s own research teams have also been testing Carbon for specialized scientific applications. These include formal code verification, enzyme assay development, and protein structure research.

In one protein research experiment, Carbon helped identify remote evolutionary relationships across large biological sequence databases. Google reports that it increased the number of homologs retrieved by an average of 40 times across seven low-homology protein families from the CASP17 blind test set. This could help improve the information available to protein structure prediction systems without requiring them to be retrained.

These results are particularly interesting because they show how advanced AI models could support scientific work that traditionally requires substantial manual investigation and specialized expertise.

Google expands access to researchers

Google is also expanding its research access initiative. The company previously opened applications for 10,000 scientists to use Gemini 4 Argon through Google Antigravity at no cost for six months.

With Carbon now available, current and future participants can also receive access to the newer model. Applications remain open, with researchers onboarded on a rolling basis, subject to eligibility, available capacity, and selection.

The program will continue providing access to Google’s latest frontier models as they become available.

Gemini 4 Carbon availability and pricing

Gemini 4 Carbon is rolling out across Google’s products for different types of users.

Developers can access the model through the paid tier of the Gemini API in Google AI Studio and Android Studio. Google AI Ultra subscribers can also use it through Google Antigravity and AI Studio.

Enterprise customers can access Carbon through Gemini Enterprise, the API, Google Antigravity, and CodeMender.

For everyday users, Gemini 4 Carbon is available through the Gemini app. Google AI Ultra subscribers can also access it through AI Mode in Google Search.

Cybersecurity defenders can apply to Google’s Fairwind Program for access to the dedicated Gemini 4 Carbon for cyber offering. Researchers can also apply for the Google Antigravity research access program.

Google is offering an introductory pricing period for the publicly available Carbon model. After November 30, 2026, the listed pricing will be $4 per million input tokens and $20 per million output tokens.

Developers looking to integrate the model into their applications can also consult Google’s updated Gemini 4 developer guide and documentation for migration instructions and recommended practices.

With Carbon joining Argon in Google’s Gemini 4 model family, the company is expanding its focus beyond general-purpose AI assistance toward models capable of handling more demanding professional and scientific workflows. Its availability across developer tools, enterprise services, research programs, and the Gemini app gives users several ways to put those capabilities to work.