Google has unveiled Gemini 4 Carbon, a frontier AI model for cybersecurity, software engineering, and long-running tasks. The model expands Google's competition with OpenAI and Anthropic, but it is not yet broadly available. (Reuters)
Carbon targets harder tasks
Google designed Carbon to handle vulnerability discovery, exploitation testing, and software patching. It also targets complex engineering projects that require many connected steps rather than short answers.
On Google's internal tests, Carbon generated secure code at more than twice the success rate of Gemini 4 Argon. Google also reported a 0.0% attack success rate on its Gray Swan prompt-injection benchmark, compared with 0.7% for Argon. These are company-reported results, not independent proof of performance. (Google)
Limited rollout protects access
Google is initially giving Carbon to a subset of cybersecurity defenders through its Fairwind Program. It plans to expand access to additional trusted partners while testing safety controls through a voluntary U.S. government process for advanced AI models.
Google said developers, businesses, and consumers will receive access through its products later, but it has not announced a public release date. The company also said it will continue gathering feedback and refining Carbon's safeguards before a wider launch. (Google)
Pricing begins with a discount
When Carbon becomes available through an API, Google plans introductory pricing of $2 per million input tokens and $10 per million output tokens. After that period, the price will rise to $4 per million input tokens and $20 per million output tokens. Cached input tokens will receive a 95% discount during the introductory period. (Google)
For now, Carbon remains limited to Google's initial cybersecurity testing program while the company prepares for broader access.
References
Google. (2026, October 28). Gemini 4 Carbon: Our next frontier model. https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
Jayaprakash, P. (2026, October 28). Google launches Gemini 4 Carbon AI model with cybersecurity focus. Reuters. https://new.streetinsider.com/Reuters/Google%2Blaunches%2BGemini%2B4%2BCarbon%2BAI%2Bmodel%2Bwith%2Bcybersecurity%2Bfocus/27248308.html
Whitwam, R. (2026, October 28). Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use. Ars Technica. https://arstechnica.com/ai/2026/10/google-announces-gemini-4-carbon-its-latest-frontier-ai-model-that-you-cant-use/