Google on Tuesday, Nov. 10, made Gemini 4 Carbon, which it calls its strongest AI model, generally available. Paying developers and businesses can use it now, and it is rolling out to Google AI Ultra subscribers in the Gemini app and AI Mode in Search. Developers pay the same introductory rate as for the Gemini 4 Argon model, $2 per million input tokens and $10 per million output tokens. Google says Carbon uses more tokens than Argon on a task because it explores and checks its work longer, so the same job can cost more.
Who should use which model
Google says Argon "remains the ideal choice for the majority of software engineering and knowledge work," and that Carbon is meant for harder jobs such as multi-step coding and debugging and open-ended research.
On the developer side, Carbon is on the paid tier of the Gemini API in Google AI Studio and Android Studio, and Ultra subscribers can try it in AI Studio and in Antigravity, Google's coding platform built around agents. Antigravity is also getting a preview feature called /swarm that splits one prompt across a team of agents.
Google pitches it on cost, too. It says Carbon ranks first on three outside benchmarks from Cursor, Perplexity and Zapier, at less than half the cost per task of the next-best model from another company.
Cybersecurity and safety
Security teams got it first. Carbon went to a group of cybersecurity defenders two weeks before the public release, through a Google program called Fairwind, and because the model is stronger at cyber tasks, Google says it added safeguards meant to keep that defensive skill from being turned to attacks.
In simulated attack tests, Google says Carbon blocked 99% of attempts by an automated red-teaming agent and cut the breach rate per attempt to 0.8%, from 3.1% on Argon. An outside tester, Trajectory Labs, found no new universal jailbreaks over 132 hours and more than 13,000 requests, according to the company. On a 375-task honesty test for agents, Carbon passed 92% of tasks, compared with 70% for Argon.
