Gemini 4 Carbon Available to AI Ultra Subscribers and Through Paid API

Google expands access to Gemini 4 Carbon beyond cybersecurity circles. The model, touted as the most powerful in the group, remains exclusive to AI Ultra subscribers and the paid API, with a gradual rollout and capabilities focused on complex tasks and agent work.

Benchmarks, positioning against Argon, and human verification

At its initial announcement, Gemini 4 Carbon scored 59 points on the Artificial Analysis Intelligence v5.0 index. Google highlights results in several programming and automation assessments, while reminding that these metrics provide benchmarks without guaranteeing performance on every task. The group also acknowledges that Gemini 4 Argon is better suited for most common professional work. Carbon, on the other hand, aims for demanding challenges, with potentially longer reasoning and increased resource consumption. The use of agent teams can speed up certain steps but multiplies the elements to control. In any case, the model's outputs require human verification.

Expanded access but reserved for Ultra subscribers and the API

On November 10, 2026, Google officially announced a broader rollout of Gemini 4 Carbon, now intended for developers, businesses, and the general public. The model, previously limited to experts through the Fairwind program, is only accessible to Google AI Ultra subscribers. Users of the free version of Gemini and AI Pro subscribers do not have access, and no date for wider availability has been communicated. Unlike Carbon, Gemini 4 Argon is available to both Pro and Ultra subscribers. Google explains that access is restricted because the model allocates more computing power to exploring and verifying its solutions. For professionals, developers can use Google AI Studio's paid API, and options exist for businesses depending on their tools.

API pricing and launch schedule

Until November 30, 2026, the public API offers an initial rate set at $2 for every million incoming tokens and $10 for every million outgoing tokens. Starting December 1, Google announces that these rates will double. These amounts pertain to the API and do not correspond to the price of the Google AI Ultra subscription. Access may also go through Google Antigravity, depending on the rights granted.

Key features: multimodal, complex tasks, and agents/swarm

Designed for advanced reasoning and cybersecurity, Gemini 4 Carbon is touted to solve complex problems and coordinate multiple agents. It accepts text, images, audio, and video as input, and is suitable for analyzing financial documents, cross-referencing information, producing reports or tables, examining scientific data, and multi-step programming. The /swarm function, offered in a preliminary version in Google Antigravity, allows AI agents to divide a project, for example, assigning architecture to one, coding to others, and verifying changes to a final agent. Google has also demonstrated examples of reconstructing 3D environments from technical documents. This method is part of its research on orchestrating agents with Gemini Enterprise and includes announced results in detecting and correcting vulnerabilities.

Where to find it: Gemini app and AI mode in Search

Google is beginning to activate Gemini 4 Carbon in the Gemini app and in the AI mode of its search engine, allowing access without going through a dedicated developer platform. Users subscribed to AI Ultra can select this model via the app selector or by using the AI mode of Google Search. As the rollout is gradual, it is possible that the option may not be available immediately. This model, which Google describes as the most powerful in its range, is aimed at professional use, programming, and research, provided the appropriate subscription is held.