Google limits Gemini 4 Carbon launch to cyber defenders

Google has announced Gemini 4 Carbon, a frontier artificial intelligence model built for cybersecurity defense and complex coding, and is initially limiting access to selected cyber defenders and trusted partners. The company said it plans to expand availability later, but has not set a public release date. Carbon follows Gemini 4 Argon by about two weeks. Google says the new model can identify software vulnerabilities, assess whether they can be exploited and generate patches. It is also designed to handle extended, multistep engineering work, including tasks that may take days, according to the company. Google reported that Carbon scored 88.2 on SECbench Pro, a test of vulnerability discovery; 54.8 on ExploitGym, which assesses exploit validation; and 71 on CWEbench, which measures patching. The company also said Carbon ranked first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59. Artificial Analysis’ evaluation identifies Carbon High as the index leader and notes that results from version 5.0 are not directly comparable with the previous version. The benchmark figures do not establish how the model will perform across real-world...