Google launches Gemini 4 Carbon for small business cybersecurity

A new standard for AI performance

Google has officially introduced Gemini 4 Carbon, its latest artificial intelligence model designed to tackle complex cybersecurity and coding challenges. The company is positioning this new release specifically to help small business owners who often face significant resource limitations. By providing advanced tools, Google hopes to help these smaller entities protect their digital assets more effectively against growing threats.

The model has already made a notable impact on industry benchmarks. According to the source, Gemini 4 Carbon currently ranks first in the updated version of the Artificial Analysis Intelligence Index. It achieved a score of 59, which reflects its high level of proficiency in various technical tasks compared to other models currently on the market.

This new release follows closely on the heels of Gemini 4 Argon, which was launched only a few weeks ago. However, Carbon is designed to outperform its predecessor in specific areas. Google officials noted that the model shows significant gains in coding, agentic tasks, and the core capabilities required for modern cybersecurity defense.

Technical benchmarks in security

To prove its effectiveness, Google tested Gemini 4 Carbon against several rigorous security evaluations. One of the most notable results came from the SEC-bench Pro tasks, where the model achieved a success rate of 88.2% in identifying vulnerabilities. This is a critical metric for businesses that need to find and fix weaknesses in their software before hackers can exploit them.

The model also performed well in other specialized testing environments. It reached a 54.8% success rate in ExploitGym tasks, which focus on the mechanics of exploitation. Additionally, it demonstrated a 71.0% success rate in patching within the CWE-bench evaluation. These numbers suggest that the model is becoming increasingly capable of not just finding problems, but also suggesting ways to fix them.

Internal benchmarks from Google further support these external results by showing improved reliability. The company indicates that Gemini 4 Carbon produces more consistent outputs in live coding environments. By reducing the rates of fabrication and misalignment, the model helps developers create more stable software with fewer delays in the production cycle.

Accessibility for small enterprises

One of the primary goals for Gemini 4 Carbon is to make high-level AI capabilities accessible to businesses that do not have massive budgets. Google is implementing a pricing structure that mirrors its previous model, Gemini 4 Argon. This includes a cost of $2 per million input tokens and $10 per million output tokens, which is intended to be competitive for smaller organizations.

To further lower the barrier to entry, Google is offering a significant discount on cached input tokens. These are priced at 95% off the standard rate, which can help businesses manage their recurring costs as they scale their use of the AI. This strategy aims to allow smaller enterprises to enhance their digital services without facing exorbitant fees.

The rollout of the technology is being handled with a strategic, phased approach. It will first be available to a select group of cyber defenders before expanding to other trusted partners through the Fairwind Program. This deliberate pace allows Google to integrate real-world feedback into the model's development to ensure it remains robust in actual security scenarios.

Implementation and human oversight

While Gemini 4 Carbon offers powerful automated tools, Google suggests that small business owners should approach the technology with a balanced perspective. Relying on machine-generated code can introduce risks if not managed correctly. Users will still need to maintain human oversight to ensure that the quality and security of the generated code meet their specific requirements.

There is also a concern regarding the potential for increased dependency on AI tools. As businesses integrate these technologies into their daily workflows, they must manage the tools carefully to avoid over-reliance. The goal is to use the AI to augment human capabilities rather than replace the need for critical thinking and strategic planning in cybersecurity.

Ultimately, the model is presented as one piece of a larger security strategy. Even with advanced AI protections, businesses are encouraged to continue investing in comprehensive measures. This includes traditional methods like user training and awareness programs. By combining Gemini 4 Carbon with human-led strategies, small businesses can create a more resilient defense against the evolving threat landscape.