Google Unveils Gemini 4 Carbon and Genie 4 With Cybersecurity Focus
Google released the Gemini 4 Carbon model on October 28, 2026, launching a specialized border model focused on cybersecurity and complex software programming just 15 days after introducing the Gemini 4 Argon version, PCGuia reported. The new system achieved an 88.2% score on the SEC-bench Pro benchmark for identifying vulnerabilities, outperforming several competing architectures in automated digital defense evaluations.
The release expands Google’s fourth-generation artificial intelligence offerings alongside the Genie 4 world-simulation model. While industry competitors typically update border systems on weekly cycles, Google took 11 months following its previous major release to deploy the Carbon architecture, according to industry reporting cited by PCGuia. The company restricted initial access to select cybersecurity professionals participating in its Fairwind program.
Cybersecurity Benchmarks and Code Generation Performance
Gemini 4 Carbon secured the top position on the Artificial Analysis Intelligence Index version 5.0 with an overall score of 59. In targeted evaluations, the system recorded 54.8% on the ExploitGym benchmark for exploit tasks and 71.0% on CWE-bench for applying code corrections.
Announcing Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities.Carbon represents a step up across most benchmarks, delivering strong frontier performance in autonomous coding and cyber defense.We’re first making Carbon… pic.twitter.com/jgrKXp3RYl
Internal evaluations conducted by Google indicated that the Carbon model generates secure code with a success rate twice as high compared to the Argon model. Furthermore, cloud security platform Wiz verified the model’s capabilities in web penetration testing through its Scan for Good initiative. During extended engineering runs, the system operates across multiple days to resolve subtle errors in massive codebases, earning 89 Elo points on the WebDev Arena to reach a 1,768 score.
Security Robustness and Government Testing Protocols
Security researchers at Grey Swan reported that the Carbon model withstood all indirect prompt injection attempts, recording a 0.0% attack success rate compared to 0.7% for the preceding Argon model. Terminal-Bench 4.0 evaluations showed fraudulent behavior success rates dropping to 0.0%, while rule-bypass attempts fell to 0.4%.
General developers and enterprise clients face a one-to-two-week waiting period before gaining API access.
Pricing Structure and Operational Costs
When general availability opens, developers will integrate Gemini 4 Carbon through API connections priced at $2 per million input tokens and $10 per million output tokens. Input tokens stored in cache receive a 95% discount to support prolonged workflows. This pricing model accompanies Google’s recent adjustments to free access tiers across its Gemini software lineup.
Concurrently, Google Cloud announced the general availability of the Gemini 4 Argon model on October 27, 2026, targeted at long-running enterprise tasks across legal, financial, and development sectors. However, benchmark results from Android Bench 2.0 published by Pplware showed that Argon achieved a 21.3% success rate on complex mobile coding tasks, trailing competing options such as Claude Sonnet 5.5 at 40% and GPT-6.1 Sol at 29.3%.
Frequently Asked Questions About Gemini 4 Carbon
Which organizations have access to Gemini 4 Carbon?
Access is currently restricted to professional cyber defenders participating in Google’s Fairwind program, with general developer and enterprise availability scheduled one to two weeks following the October 28, 2026 launch.
What distinguishes Gemini 4 Carbon from Gemini 4 Argon?
While Argon is designed for general enterprise workflows and has maintained a public endpoint since October 13, 2026, Carbon is a specialized border model engineered specifically for cybersecurity analysis, vulnerability patching, and complex software programming.
How much will it cost developers to use the Carbon model via API?
Google established an introductory API price of $2 per million input tokens and $10 per million output tokens, with cached input tokens qualifying for a 95% discount.
What were the results of third-party safety testing on Carbon?
Grey Swan security researchers reported that Carbon resisted all indirect prompt injection attempts with a 0.0% attack success rate, improving upon the 0.7% rate recorded by the Argon model.