Google Unveils Gemini 4 Carbon AI Model for Cybersecurity

Google Expands AI Portfolio with Cybersecurity Focus

Alphabet’s Google has officially introduced Gemini 4 Carbon, its latest frontier artificial intelligence model, marking a strategic escalation in the competitive landscape of AI development. The launch arrives just two weeks after the introduction of Gemini 4 Argon, signaling a rapid expansion of the Gemini family tailored for developers and enterprise customers. This new model specifically targets the intensifying race to build more capable coding and cybersecurity systems, positioning Google directly against rivals like OpenAI and Anthropic.

Performance and Benchmarking

According to Google, Gemini 4 Carbon delivers state-of-the-art performance across critical cybersecurity tasks, including vulnerability discovery, exploitation analysis, and software patching. Internal benchmarks indicate that the model generates secure code at more than twice the success rate of Gemini 4 Argon. The architecture is specifically designed to handle complex, long-running engineering and coding tasks, addressing a key need in modern software development workflows.

Koray Kavukcuoglu, Senior Vice President at Google DeepMind and chief AI architect at Google, described Gemini 4 Carbon as the company's strongest model to date. He emphasized its particular strength in cybersecurity applications and its ability to manage sustained, complex computational tasks. This endorsement underscores Google's commitment to leveraging its DeepMind subsidiary to maintain technological leadership in specialized AI domains.

Market Position and Competition

The introduction of Gemini 4 Carbon places Google in direct competition with Anthropic’s Claude Opus 5.5. The pricing strategy reflects this competitive dynamic, with Carbon initially priced at $2 per million input tokens and $10 per million output tokens. This introductory pricing matches that of Gemini 4 Argon. Once the introductory period expires, Google plans to raise prices to $4 per million input tokens and $20 per million output tokens, a tier that aligns with the pricing of Claude Opus 5.5.

Performance metrics further illustrate the model's standing. Gemini 4 Carbon has taken the lead in SEC-bench Pro, ExploitGym, and CWE-bench benchmarks. On the WebDev Arena, it secured fourth place, trailing behind Claude Opus 5.5, GPT-6.1 Astra, and Sonnet 5.5. These results highlight both its strengths in security-specific tasks and its competitive position in general web development scenarios.

Availability and Future Pricing

Access to Gemini 4 Carbon is currently restricted. The model is being made available first to a subset of cybersecurity defenders through Google’s Fairwind Program. This limited rollout allows Google to continue testing safeguards and monitoring performance in real-world scenarios before a broader release. The company has not specified the duration of the introductory pricing period, noting only that it is indeterminate.

The irony of the name "Carbon" has not been lost on observers, particularly given Google's stated goals for a carbon-dioxide-free future. While some had speculated on alternative names like "Americium" or "Silicon" to reflect management's worldview or technological ambitions, the chosen name stands as a distinct marker in Google's periodic table-inspired model naming convention.

Strategic Implications

This launch follows a period of uncertainty for Google's DeepMind subsidiary following recent leadership changes. The impressive benchmark scores and strategic focus of Gemini 4 Carbon suggest that the subsidiary remains a potent force in AI research and development. By targeting the high-value cybersecurity sector, Google aims to capture enterprise customers who require robust, secure AI tools for their infrastructure.

The move also reflects the broader industry trend of specialized AI models. Rather than relying solely on general-purpose models, companies are increasingly developing tools optimized for specific, high-stakes tasks. Gemini 4 Carbon’s focus on vulnerability discovery and secure code generation addresses a critical need in an era of increasing cyber threats.

Key takeaways

• Google launched Gemini 4 Carbon, an AI model optimized for cybersecurity and complex coding tasks.
• The model is initially priced at $2 per million input tokens and $10 per million output tokens.
• Gemini 4 Carbon leads in SEC-bench Pro, ExploitGym, and CWE-bench benchmarks.
• The model is available via the Fairwind Program for a subset of cybersecurity defenders.
• Prices will increase to $4 and $20 per million tokens after the introductory period.
• The launch intensifies competition with Anthropic's Claude Opus 5.5 in the enterprise AI market.