Google has announced the latest addition to its family of advanced AI models; Gemini 4 Carbon. The new model was announced on October 28th 2026 and promises to help improve software development, complex engineering tasks, plus improve cybersecurity defences.
Gemini 4 Carbon offers significant improvements in identifying, testing and patching vulnerabilities within software. This shows just how much Google is focusing on using AI within cybersecurity.
This follows the announcement of Gemini 4 Argon.
Gemini 4 Carbon Can Help Improve Cybersecurity
One of the key functions of Gemini 4 Carbon is assisting those in cybersecurity.
There are now very complex vulnerabilities within modern software systems that can pose a risk to businesses, government organisations and individuals. Finding these issues and creating solutions often takes lots of knowledge and time.
Gemini 4 Carbon helps security researchers identify flaws, determine if they are exploitable and create patches for software.
The model achieved an impressive score of 88.2% on the SEC-bench Pro security benchmark.
It also scored 54.8% on the ExploitGym benchmark which assesses the ability to exploit vulnerabilities and 71.0% on the CWE-bench which tests the ability to create fixes for software.
This puts Gemini 4 Carbon ahead of several models.
Gemini 4 Carbon Can Help Improve Secure Coding
Outside of helping with cybersecurity research, Gemini 4 Carbon is being promoted as a great tool for developers.
It can assist with complex coding tasks.
Google says Carbon can achieve over double the success rate of Gemini 4 Argon for secure coding.
The company also mentions that during internal testing, the Chrome team saw Carbon perform over 10 times better than another frontier model from one of its competitors whilst creating high quality patches.
This could prove to be a valuable tool to give developers an AI assistant to help with large scale software development, debugging complex systems and improving application security.
However, this doesn’t mean it will perform the same in all situations.
Google Further Boosts Rivalry With OpenAI and Anthropic
With the launch of Gemini 4 Carbon we see an increasing rivalry between the big names in AI.
Google is competing with companies like OpenAI and Anthropic to create more powerful AI models for businesses, developers, researchers and everyday users.
Following the release of Gemini 4 Argon, Google are trying to push ahead with development and release more powerful models.
As mentioned, cybersecurity is becoming an increasingly important factor in the battle between the biggest names. Advanced AI models can help identify vulnerabilities quicker but can also cause new risks.
Google says the model has been developed with improved safety, security and protection from malicious prompts.
During testing, Carbon had a 0% success rate when being attacked using indirect prompt injection. It also made around half as many unsupported claims than Gemini 4 Argon.
These results come from testing by Google so don’t mean it will be completely safe from any risks.
Gemini 4 Carbon Will Only Be Available To A Selected Few
Whilst the new model has been announced, it won’t be immediately available to everyone.
Google are taking a slow approach and will only release it to a select few cybersecurity defenders.
They will then release it to more people using their Fairwind Program which links trustworthy organisations working on cybersecurity and protecting digital infrastructures.
This helps Google understand what the model can do, collect feedback and ensure security is improved.
The company have also mentioned they are involved in the US Governments voluntary testing program for advanced AI models.
Developers, businesses and everyday consumers will have to wait to use it.
Gemini 4 Carbon Pricing
Google have announced that Gemini 4 Carbon will be released at the same price point as Gemini 4 Argon.
The price will be;
$2 per million input tokens.
$10 per million output tokens.
After the initial release price will be increased to $4 per million input tokens and $20 per million output tokens.
This will help developers and organisations looking to include advanced AI features into their software.
However, this doesn’t mean it will be immediately available to all developers.
The Implications Of Gemini 4 Carbon
The release of Gemini 4 Carbon shows how AI is no longer just used for basic chatbots and content creation.
Advanced AI models are now being created to help with complex tasks, analysing software, finding vulnerabilities and assisting engineers.
This could help businesses save time when analysing software vulnerabilities and help improve application security.
For those in cybersecurity, having an AI assistant to help find vulnerabilities and create patches could also be a valuable addition to their security.
However, more advanced AI models can create problems. Whilst they can be useful for researching vulnerabilities they could also be abused. It will be important to responsibly implement them and ensure there is human oversight along with strong security.
Google are taking a cautious approach and are only releasing to a select few in cybersecurity.
Conclusion
Overall, Gemini 4 Carbon shows another step forward with Google’s AI strategy. The model has improvements in finding vulnerabilities, secure coding, creating patches for software and complex engineering tasks. This should help improve AI cybersecurity.
The slow release shows Google are cautious with offering such powerful AI.
As the rivalry with OpenAI and Anthropic continues, this could prove to be an important model for Google to increase their presence within enterprise AI, software development and cybersecurity.
We will need to see if it can deliver the same results in real world situations.
