Google's Gemini 4 Carbon is designed for multi-step AI work.
The company says it can identify and verify serious security flaws, then generate fixes; Google reports Carbon outperformed Gemini 4 Argon, GPT-6 Astra and Claude Opus 5.5 in tests.
Initial access is limited to specialists working on cyber defense.
According to Google, Carbon outperformed Gemini 4 Argon, OpenAI’s GPT-6 Astra, and Anthropic’s Claude Opus 5.5 in tests for programming, information processing, science, math, understanding different types of content, and digital security. The company says the model can identify serious security vulnerabilities, verify whether they are real, and generate high-quality fixes.
Google also says that Carbon effectively resists attempts to bypass its security rules. In internal testing, such attempts succeeded at a rate of 0,0% for Carbon, 0,7% for Argon, and 1,0% for Claude Opus 5.5. The rates are for Google's specific tests.
Initially, Carbon will be available to a subset of experts working on defense against cyberattacks. The initial fee is $2 per million data units the model accepts and $10 per million units it produces. Google will gradually expand access to trusted partners through the Fairwind program. It also says it wants to make the model available to developers, businesses, and consumers as soon as possible.
G
HACKS.GR / GOOGLE SEARCH
Hacks.gr on Google
Set Hacks.gr as a preferred source.
Preferred source
↗
end of article
Top comments (0)