Google Introduces Gemini 4 Carbon Frontier AI Model with Advanced Cybersecurity Capabilities

In brief: Google has launched Gemini 4 Carbon, a frontier AI model focusing on cybersecurity, coding, and extended multi-step agentic tasks.

Google Introduces Gemini 4 Carbon Frontier AI Model with Advanced Cybersecurity Capabilities

Google has officially introduced its newest frontier artificial intelligence system, designated Gemini 4 Carbon. The system brings notable improvements in several technical areas, with a primary emphasis on cybersecurity defense, software coding, and complex agentic tasks.

Check Latest Offers

Performance and Benchmark Rankings

According to the initial rollout details, the newly released system achieves top placement on the Artificial Analysis Intelligence Index, specifically version 5.0. In this index, it ranks ahead of competitors such as Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra. Furthermore, the model leads on the SEC-bench Pro and ExploitGym evaluation benchmarks, demonstrating that it outperforms top models from rival organizations on specific performance tests.

The Carbon release builds directly upon the Argon model that was made available earlier in the same month. Among its improvements, the new system reduces unsupported claims by a factor of two compared to Argon, while also showing enhanced resistance against prompt injection attacks and a lower tendency to take computational shortcuts.

Engineering and Extended Task Execution

Engineered to handle prolonged operational workloads, the model can execute continuously for days when managing multi-step engineering assignments. This capability supports complex coding and agentic workflows that require sustained processing over extended periods.

Access and Availability

Immediate access to the model is restricted. It is currently available exclusively to a carefully selected group of cybersecurity defense companies. Google plans to broaden access eventually to other participants in the Fairwind Program. However, a specific timeline detailing when this expansion will take place has not been provided.

Pricing Structure

For organizations and developers eligible to utilize the system, the token-based pricing structure is set at $2 per million input tokens. Output tokens are priced at $10 per million.

Conclusion

Google's release of Gemini 4 Carbon marks an expansion of its AI offerings, prioritizing specialized defensive cybersecurity tools, enhanced coding reliability, and long-running agentic operations ahead of broader partner availability.

Source for this report: theverge.com.

Previous Post Next Post