Google is expanding access to Gemini 4 Carbon, its latest AI model, after first releasing it to a select group of “cybersecurity defenders” two weeks ago. The company describes Carbon as its strongest model yet and says it is designed for tasks that push beyond routine software engineering and knowledge work.
A model aimed at demanding tasks
Google says Carbon “pushes the frontier even further” than Argon, another frontier model the company launched only last month. Google expects Argon to remain the ideal choice for most software engineering and knowledge work, while positioning Carbon for work that tests the limits of a model’s intelligence.
The examples Google gives include complex, multi-step coding and open-ended scientific research. Both involve more than producing a quick answer: they can require a model to work through a longer sequence of decisions or explore a question without a narrowly defined endpoint.
That distinction suggests Carbon is being presented as a specialized option within Google’s model lineup. Argon is described as suitable for the majority of common professional tasks, while Carbon is meant for users who need more capability on particularly challenging work.
Antigravity previews coordinated agents
Alongside the model rollout, Google is previewing a “swarm” feature on Antigravity, its agentic development platform. The feature is intended to turn a single prompt “into a coordinated team of agents.”
That approach is aimed at long-horizon work, where a task may call for multiple steps and agents working in coordination. The announcement connects the feature to Carbon’s potential usefulness for agent-based work, though it does not describe how the agents divide tasks or how users will direct them beyond the initial prompt.
For developers, the pairing of a model built for complex tasks with a team-based agent feature points to a broader way of working: users can describe an objective, while AI agents handle parts of the work together. The practical value will depend on how well the system coordinates those agents on real projects.
Where Carbon is available and what it costs
Google says Carbon is rolling out through the Gemini API in Google AI Studio and Gemini for business for all paying developers and enterprises. Google AI Ultra subscribers can also use it in the Gemini app, AI Mode in Search, and Antigravity.
The company has not made clear when, or whether, the model will reach lower-tier subscribers or free users. For now, access depends on having one of the paid plans or developer and enterprise access listed in the announcement.
At launch, the introductory price is $2 per million input tokens and $10 per million output tokens. Those prices are set to double at the end November, when the introductory period ends. Google says Carbon will remain significantly cheaper than frontier rivals on a per-token basis after the increase.
Per-token pricing does not by itself determine the total cost of a task. Different models may use different numbers of tokens to complete the same work, so a lower rate for each token does not necessarily mean a lower bill in every case.
Benchmarks and the competitive picture
Benchmarks Google cited when it first announced Carbon suggest it performs among the most powerful models across most metrics, behind Fable 5.2. Those results offer one view of the model’s capabilities, but benchmark comparisons should be treated carefully: companies can select results that present their own products in a favorable light.
Carbon’s launch comes soon after Argon and signals Google’s return to the leading edge of frontier AI development, according to the source article. It also adds another contender to a field that had appeared for months to be dominated by Anthropic and OpenAI.
For prospective users, the announcement leaves two practical questions: whether Carbon’s performance suits their specific work, and how its token use affects the cost of that work. The new model, its pricing, and Antigravity’s agent swarm give developers more options to assess as Google broadens access.