Google launches Gemini 4 Carbon with major cybersecurity gains
Google announced Gemini 4 Carbon, its new frontier AI model featuring major improvements in cybersecurity, coding, and multi-step agentic tasks.
On Wednesday, Google announced the release of Gemini 4 Carbon, its latest frontier artificial intelligence model. Following shortly after the release of the Argon model earlier this month, Carbon delivers significant performance gains in coding, scientific analysis, and agentic tasks—which are AI programs designed to complete complex, multi-step workflows autonomously.
Google reports that Gemini 4 Carbon outperforms top competing models from OpenAI and Anthropic on several industry benchmarks, including the Artificial Analysis Intelligence Index. The new model is currently available only to select cybersecurity defense companies and partners in Google’s Fairwind Program due to its advanced capabilities.
What are the capabilities of Gemini 4 Carbon?
Google highlights a major leap in cyberdefense capabilities for Gemini 4 Carbon. The model is designed to find software vulnerabilities, validate them, and generate secure code with fewer flaws than its predecessor, Argon. According to Google, Carbon is twice as likely to generate vulnerability-free code compared to Argon.
The model is also built to handle longer, more complex tasks. Internal testing showed Carbon running for days through multi-step engineering assignments without giving up. Additionally, the Chrome team noted that Carbon excels at producing high-quality software patches compared to other available models.
What are the safety and security features?
Alongside its technical improvements, Gemini 4 Carbon features enhanced safety measures. Google reports that the model is better at avoiding shortcuts and resisting prompt injection attacks, which are malicious attempts to trick an AI into bypassing its rules. In specific safety evaluations, Carbon successfully resisted all attempted prompt-injection attacks.
Google also states that Carbon offers greater transparency by reducing unsupported claims by half compared to Argon.
When can you use it and what does it cost?
A wide public release date for Gemini 4 Carbon has not been announced. Google is taking a deliberative approach to releasing the model, following voluntary testing frameworks established by the United States government. Once initial testing concludes, it will roll out more broadly through the Fairwind cyber defense program, and eventually to AI Ultra subscribers and developers using the Gemini API.
When available via the API, Carbon will launch with an introductory price of $2 per million input tokens and $10 per million output tokens, matching the introductory rate set for the Argon model.
Comments 0
No comments yet. Start the conversation.