Gemini 4 Carbon is now available to developers and consumers
Google has opened Gemini 4 Carbon to paying developers and enterprises, while rolling it out to Google AI Ultra subscribers across its consumer products.
Google has made Gemini 4 Carbon available to developers, enterprises, scientists and consumers, moving the model beyond its earlier limited cybersecurity preview.1
Carbon is aimed at difficult, multi-step work rather than quick everyday prompts. Google says it is designed for complex coding and debugging, open-ended scientific research, cybersecurity and long-horizon agentic workflows. The company describes it as its strongest model, although those performance claims remain Google's own assessment.1
The model is available through the API in Google AI Studio and Gemini for business for paying developers and enterprises. Google is also rolling it out to Google AI Ultra subscribers through Antigravity, the Gemini app and AI Mode in Search.1
Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, matching the launch pricing for Gemini 4 Argon. The company says Carbon uses more tokens than Argon to explore problems thoroughly, reason through alternatives and run extended verification loops, so the headline token price does not necessarily describe the final cost of a task.1
Google is previewing a /swarm feature in Antigravity alongside the launch. It lets developers assign Carbon to demanding roles such as architectural design, coordination and auditing, while pairing it with faster models and isolated workspaces. The Google blog's accompanying demonstration shows agents reconstructing the Eiffel Tower from historical photographs and engineering plates.1
The wider availability is the meaningful change here: Carbon is no longer only a model for specialist preview programmes, but Google is still separating access by product and subscription tier. Free or lower-tier availability has not been announced, and Google's benchmark and cost comparisons should be read as company-reported results rather than independent guarantees.
Our opinion
Gemini 4 Carbon looks most relevant as a high-end tool for work where checking the answer matters as much as producing it. The pricing is competitive on paper, but the model's appetite for extra reasoning tokens means developers will need real workload data before assuming it is cheaper. The useful test will be whether Carbon's deeper loops reduce human correction time, not whether it wins a single benchmark chart.