Gemini 4 Carbon Rolls Out to AI Ultra Subscribers and Search AI Mode

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google’s new AI model for cybersecurity and long-horizon agentic work, including knowledge work, software engineering, and science. Google made the model available to paying developers and enterprises on November 10, 2026, through Google AI Studio and Gemini for business. Google’s Gemini 4 Carbon announcement describes the expansion from a limited cyber-defender release to broader paid access.

Google describes Carbon as its strongest model for cybersecurity and extended agentic workflows. Agentic workflows are tasks where an AI system works through several connected actions rather than returning one answer to one prompt. That makes Carbon relevant to teams that need help analyzing large sets of materials, planning work, or assisting with technical tasks, but the model’s output still requires human oversight before it affects production systems or security decisions.

Gemini 4 Carbon’s cybersecurity positioning also makes careful review more important. Security work can involve code, credentials, infrastructure details, and incident information, so users should keep sensitive material within approved organizational policies. Readers evaluating AI tools for security work can use core cybersecurity concepts to distinguish between general assistance and a verified security control.

Where can Google AI Ultra subscribers use Gemini 4 Carbon?

Google AI Ultra subscribers can receive Gemini 4 Carbon in Google Antigravity, the Gemini app, and AI Mode in Search. Google announced the rollout on November 10, 2026, alongside access for paying developers and enterprise users. The availability change gives paid consumer subscribers more places to use the model, rather than limiting it to an API or a developer product.

Gemini 4 Carbon’s placement across these surfaces matters because each product supports a different type of task. AI Mode in Search is designed around information-seeking prompts, while the Gemini app is a broader assistant interface and Antigravity is where Google is previewing coordinated agent workflows. Independent coverage of the November 10 availability expansion is available from Engadget’s rollout report.

Google has described the consumer rollout as a rollout, not a guarantee that every eligible account sees Carbon immediately. Subscribers should check the model selector and product interface for availability rather than assuming every Google AI Ultra account has the same options at the same time. Search users should also remember that an AI Mode answer is generated output, not a substitute for opening and evaluating original sources.

How much does Gemini 4 Carbon cost through the API?

Gemini 4 Carbon’s introductory API price is $2 per 1 million input tokens and $10 per 1 million output tokens through November 30, 2026. Google says the price rises after that date to $4 per 1 million input tokens and $20 per 1 million output tokens. The lower introductory rate gives developers a short period to test workloads, but teams should budget using the post-November pricing if the model will remain in production.

Input tokens are the text, files, instructions, and other material sent to the model. Output tokens are the material the model produces in response. A large context window can be useful for long documents and multimodal analysis, but it can also increase costs when users repeatedly send extensive material or request long generated responses.

Gemini 4 Carbon pricing should be evaluated alongside the amount of material a workflow sends and receives, not only the number of prompts. A single agent task can use substantially more tokens than a short chat because it may read source material, reason through intermediate steps, and generate a detailed result. The practical response is to monitor token use during a small pilot before expanding access across a team.

What can Gemini 4 Carbon process?

Gemini 4 Carbon accepts text, images, audio, and video, with a combined input-output context window of up to 1 million tokens. Google DeepMind lists those capabilities in the model card, which also documents the model’s known limitations. Google DeepMind’s Gemini 4 Carbon model card is the most useful source for understanding what the model is designed to accept and where Google identifies residual risk.

A 1 million-token context window means Carbon can take in and generate a large combined amount of material during a task. That capacity can support work involving long reports, multiple files, recordings, images, or video, but a larger context does not guarantee that every detail is interpreted correctly. Users should still isolate critical claims, calculations, and security findings for direct verification.

Gemini 4 Carbon’s multimodal design may be particularly useful when a task combines visual and written evidence, such as reviewing screenshots alongside documentation. At the same time, users should avoid uploading confidential files without confirming their organization’s approved AI policy. The broader debate over AI training controls also shows why teams should understand how a service handles submitted information before using it for sensitive work.

What limits and risks does Gemini 4 Carbon have?

Gemini 4 Carbon can hallucinate, experience occasional slowness or timeouts, and use more tokens at higher thinking levels. Google identifies those limitations in the Carbon model card. Hallucinations are incorrect or unsupported AI-generated claims, which means a fluent answer can still contain a false conclusion, fabricated citation, or mistaken interpretation of supplied material.

Gemini 4 Carbon’s higher token use at higher thinking levels matters because a more elaborate reasoning setting can affect both cost and response time. A timeout also creates a workflow risk when an application expects an answer within a fixed period. Developers should design systems that can retry safely, record failures, and route high-impact decisions to a person instead of treating one generated response as final.

Gemini 4 Carbon should not be used as the sole authority for security remediation, legal interpretation, medical guidance, or financial decisions. The model can help summarize, organize, and propose next steps, but a qualified professional or responsible system owner should verify the result before action is taken. The most sensible approach is to use Carbon for acceleration and analysis while keeping evidence review and approval with a human.

What is the Gemini 4 Carbon /swarm feature?

Gemini 4 Carbon is connected to a previewed Antigravity feature called /swarm, which Google says turns one prompt into a coordinated group of agents that users can monitor in real time. The feature is intended to divide a broader request among multiple AI agents, rather than asking one model instance to perform every part of the work in isolation.

A coordinated-agent setup can make complex tasks easier to break into research, planning, coding, review, and other stages. That said, more agents do not automatically make a result more accurate. A workflow can repeat an early error across multiple steps if users do not set clear instructions, check intermediate results, and define which actions require approval.

Gemini 4 Carbon’s /swarm preview is most relevant to users who already understand agent workflows and can monitor them closely. Readers who are new to AI-driven coding should understand the risks behind vibe coding practices, especially when generated code or automation could affect real accounts, files, or services. Users should start with reversible, low-risk tasks before allowing any agent group to work with important data.

Who should use Gemini 4 Carbon?

Gemini 4 Carbon is best suited to paying developers, enterprises, and Google AI Ultra subscribers who need a model for complex multimodal work or long-running agentic tasks. Google specifically positions Carbon around cybersecurity, knowledge work, software engineering, and science. Those areas often involve many interconnected documents, instructions, and decisions, which is where a large context window can be useful.

Gemini 4 Carbon is less suitable for users who need guaranteed accuracy, instant responses, or predictable low token consumption for every task. Google’s stated limitations mean the model can return an incorrect answer, take longer than expected, or consume more tokens under higher thinking settings. For most users, a smaller or less complex model may remain the more practical choice for routine questions.

Google AI Ultra subscribers should first confirm where Carbon appears in their account, then test it with non-sensitive work that has a clear correct answer. Developers should measure input and output token use under the current introductory price and again plan for the higher rate after November 30. Organizations should establish review rules before using Carbon in workflows involving code deployment, security findings, or confidential records.

FAQ

Is Gemini 4 Carbon available to Google AI Ultra subscribers?

Yes, Gemini 4 Carbon is rolling out to Google AI Ultra subscribers through Google Antigravity, the Gemini app, and AI Mode in Search. Eligible subscribers should check the relevant product interface because rollout availability can vary by account and surface.

How much does Gemini 4 Carbon cost?

Gemini 4 Carbon costs $2 per 1 million input tokens and $10 per 1 million output tokens through November 30, 2026. Google says the API price increases after that date to $4 for input tokens and $20 for output tokens per 1 million tokens.

Can Gemini 4 Carbon analyze images, audio, and video?

Yes, Gemini 4 Carbon accepts text, images, audio, and video. Google lists a combined input-output context window of up to 1 million tokens, but users should verify important details in multimodal results.

Does Gemini 4 Carbon have known limitations?

Yes, Gemini 4 Carbon can hallucinate, encounter occasional slowness or timeouts, and use more tokens at higher thinking levels. Users should review important output before using it for security, financial, medical, legal, or production decisions.

What does Gemini 4 Carbon /swarm do?

Gemini 4 Carbon /swarm is a previewed Antigravity feature that turns one prompt into a coordinated group of agents users can monitor in real time. Users should begin with low-risk tasks because coordinated agents can still repeat errors without human review.