Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity

THE NEWS

What happened

Google has launched Gemini 4 Carbon, a new frontier model specifically optimized for cybersecurity defense. It currently outperforms leading models from OpenAI and Anthropic on critical benchmarks like SEC-bench Pro and ExploitGym, excelling at finding vulnerabilities and generating secure code. While initially restricted to select partners in the Fairwind Program, the model promises significant improvements in handling complex multi-step engineering tasks and resisting prompt injection attacks.

CONTEXT

Why it matters

Google has officially launched Gemini 4 Carbon, marking a significant leap in AI-driven cybersecurity. This new frontier model currently tops the Artificial Analysis Intelligence Index (v5.0), surpassing leading competitors from OpenAI and Anthropic. Specialized for finding vulnerabilities and generating secure code, it excels on critical benchmarks like SEC-bench Pro and ExploitGym. While initially available only to select partners in Google's Fairwind Program, the model offers enhanced transparency and the ability to tackle complex, multi-step engineering tasks that persist for days. Priced competitively at $2 per million input tokens and $10 per million output tokens, Gemini 4 Carbon represents a major shift in how AI is deployed for enterprise defense.

AT A GLANCE

Key facts

• Gemini 4 Carbon holds the top spot on the Artificial Analysis Intelligence Index (v5.0), ahead of Claude Opus 5.5 and GPT-6 Astra.
• The model currently outperforms competitors on SEC-bench Pro for discovering real-world bugs and ExploitGym for turning vulnerabilities into exploits.
• It is an upgrade to the previous Argon model, offering improved transparency by reducing unsupported claims by two times.
• The model is designed to run for days to persist through complex engineering tasks without taking shortcuts.
• Pricing matches the previous Argon model at $2 per million input tokens and $10 per million output tokens.
• Access is currently limited to a select group of cybersecurity defense companies before expanding via the Fairwind Program.