Google DeepMind releases Gemini 4 Carbon frontier model

Google DeepMind announced Gemini 4 Carbon on October 28, 2026, positioning it as the company's newest frontier model with cybersecurity and long horizon capabilities. The release was also reported by blog.google, The Decoder, Mashable, and VentureBeat. The Decoder noted that the release arrived just two weeks after Google introduced its Gemini 4 Argon model. [1][2][3][5][6][11]

According to DeepMind Research News, Gemini 4 Carbon reached the top spot on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. VentureBeat similarly reported that the model claimed the number one ranking on the updated benchmark index. As FTMQ SI reported earlier, Carbon took the top position ahead of rival models Claude Opus 5.5 and GPT-6 Astra. [1][6][11]

DeepMind Research News reported specific performance figures across several cybersecurity evaluation suites. Gemini 4 Carbon achieved 88.2 percent on SEC-bench Pro for identifying software vulnerabilities, 54.8 percent on ExploitGym for exploitation tasks, and 71.0 percent on CWE-bench for patching code. [1]

The Gemini family of multimodal models was first announced by Google DeepMind in December 2023 as a successor to LaMDA and PaLM 2. As FTMQ SI reported earlier, internal tests show that Gemini 4 Carbon doubles previous performance in generating secure code. [7][9][11]