Google launches Gemini 4 Carbon with a significant leap in cybersecurity

Google has released Gemini 4 Carbon, its latest frontier model, with significant gains in coding, agentic tasks, and cybersecurity. The company says the model brings a significant leap in cyberdefense, and it currently leads several benchmarks mentioned in the announcement. It is available first to a select group of cybersecurity defense companies, with later expansion planned for other partners in the Fairwind Program.

Benchmark results

Google highlights Carbon's top spot on the Artificial Analysis Intelligence Index (v5.0), with Claude Opus 5.5 and GPT-6 Astra trailing behind. Carbon also holds the number one spot on SEC-bench Pro, which measures an AI model's ability to discover real-world bugs, and on ExploitGym, which measures how a model can turn a vulnerability into exploits. Google says Gemini 4 Carbon also beat GPT-6 Astra on SEC-bench Pro.

Cybersecurity and task performance

According to Google, Carbon is better at finding vulnerabilities and generating secure code. The company says it is built for completing longer tasks, and its internal teams have seen it run for days to persist through dead ends to tackle complex, multi-step engineering tasks. Google also says Carbon is better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users.

Transparency and pricing

Google says Carbon offers more transparency and honesty by cutting unsupported claims by two times when compared to Argon, which Google released earlier this month. It will launch at the same competitive introductory price as Argon, at $2 per million input tokens and $10 per million output tokens. That rate doubles once the introductory period ends.

Sources: The Verge · Ars Technica