Google Opens Gemini 4 Carbon to Developers and Businesses After Cyber-Only Preview

Paying developers and enterprises can use Carbon through the Gemini API in Google AI Studio and through Gemini for business. The model is also available in Android Studio and CodeMender, Koray Kavukcuoglu, Google DeepMind's senior vice president and Google's chief AI architect, wrote in a company blog post. Google AI Ultra subscribers are getting Carbon in the Gemini app, in AI Mode in Search and in the Antigravity coding tool, with that rollout under way.

For corporate buyers, the main question is cost. Until November 30, Carbon is priced at $2 per million input tokens and $10 per million output tokens, the same as the existing Gemini 4 Argon. After that the price doubles to $4 and $20. Google also says Carbon uses more tokens than Argon because it spends longer exploring and checking its work, so a task can cost more on Carbon even when the per-token price is the same.

Google is not pitching Carbon as Argon's successor. "Argon remains the ideal choice for the majority of software engineering and knowledge work," the company said. It is steering Carbon towards complex multi-step debugging and open-ended scientific research. In the Gemini app, 9to5Google reported, the model picker labels Argon "Code & analysis" and Carbon "Most complex work".

Benchmarks Against Fable 5.2

Google said Carbon ranked first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at less than half the cost per task of the next-best outside model. It said Carbon placed second on a new Multimodal Knowledge Agent benchmark, built with Artificial Analysis, at less than a sixth of the leader's average cost per task. The Decoder identified that leader as Anthropic's Claude Fable 5.2, released a day earlier. Fable 5.2 also sits two points above Carbon on Artificial Analysis's overall index, 62 to 60, when both run at comparable settings.

The Decoder put Carbon's cost on Artificial Analysis's tests at $7.76 per solved task at standard pricing. Fable 5.2 cost about $9 per task at its High setting and $26.50 at its maximum.

What the Cyber Preview Found

Google published results from the two-week preview. When Carbon scanned two major software-as-a-service providers for security firm Wiz's Scan for Good programme, it found previously unknown vulnerabilities in under two hours. Google's Cloud Vulnerability Research team used the model to show that two bugs in popular open-source software could be exploited, and the maintainers reclassified both as critical and patched them. In Project Fortify, Google's automated system for scanning, triaging and fixing flaws, Carbon reached 100% recall on scanning and 95% precision on triage in an internal benchmark.

Google said its leading scores on the SEC-bench Pro, ExploitGym and CWE-bench v1 security benchmarks apply to the Carbon version offered through Fairwind, and that cybersecurity-specific access still runs through that programme. In simulated attacks by an adaptive red-teaming agent, Carbon blocked 99% of attack trajectories, and the rate at which individual attempts succeeded fell to 0.8%, from 3.1% for Argon. Google also said prompt injection robustness remains an area for improvement.

Alongside the release, Google is previewing a /swarm command in Antigravity that turns a single prompt into a coordinated team of agents. Users can assign each agent a different model, so Carbon can handle planning, coordination and review while lighter models work in parallel and check its output.

Google is also extending Carbon to scientists in its programme offering 10,000 researchers six months of free Antigravity access, and applications remain open. The introductory API price ends on November 30.