Google Unveils Gemini 4 Carbon, Opens It to Cyber Defenders First

Google has announced Gemini 4 Carbon, its newest frontier AI model. The company says it delivers significant gains in coding, agentic tasks and cybersecurity defense. The announcement comes just two weeks after the launch of Argon.

According to Google, Carbon ranks first on the recently updated 5.0 version of the Artificial Analysis Intelligence Index, with a score of 59. The model is said to be coming soon for developers, businesses and consumers, but Google has not given a firm date for general availability.

Where Carbon improves

Google says the most visible progress falls under three headings: coding, agentic tasks and cyber defense. On the defense side, it particularly highlights the model's ability to find security vulnerabilities.

Agentic tasks refer to an AI model going beyond answering a single question to plan and carry out multi-step work on its own. Google describes its gains here as significant progress, but the announcement includes no detailed comparison or concrete use case.

Building on user feedback after the previous launch, the company also says it improved the model for scientific tasks, end-to-end web development and multimodal understanding, meaning the ability to process different kinds of data, such as images, alongside text.

Why releases are coming so fast

Google attributes the short gap between the two models to researchers and models working together to advance capabilities more quickly. In other words, the company says its research teams and its models are progressing hand in hand, and that this is shrinking the interval between releases.

Safety work and access plans

On safety, Google says it focused on improving the model's alignment, and that Carbon was trained to respond appropriately to complex or high-pressure tasks.

As for access, Carbon will first be offered to cyber defenders this week. For broader availability, Google used only the word soon, giving no specific date for developers, businesses or consumers.

Pricing

When the model becomes available, Google will apply a special introductory price: $2 per million input tokens and $10 per million output tokens, with a 95 percent discount on the input price for cached input tokens.

After the introductory period ends, prices will rise to $4 per million input tokens and $20 per million output tokens. Tokens are the small units AI models use when processing text, and API pricing is usually calculated on them. Google has not said how long the introductory period will last.

Overall, Google has made its second major model announcement in a short span. Carbon's strong position in coding, agentic tasks and cyber defense rests on the company's own claims until its real-world performance becomes clear, and the general availability timeline and the length of the introductory pricing remain uncertain.