Google releases Gemini 4 Carbon with cybersecurity and coding focus

Google released its new frontier model, Gemini 4 Carbon, on October 28. Mashable, SiliconANGLE AI, and The Verge all reported that the rollout comes two weeks after Google released its previous flagship model, Gemini 4 Argon. [1][2][3]

In a company blog post, Google DeepMind senior vice president Koray Kavukcuoglu called Carbon the company's strongest model yet, according to Mashable. Mashable noted that Kavukcuoglu leads Google DeepMind following the departure of former chief executive Demis Hassabis. [1]

According to SiliconANGLE AI, internal tests show Carbon is more than twice as effective at generating secure code compared with prior performance. Mashable and The Verge reported that the model excels at long-horizon jobs and agentic tasks. SiliconANGLE AI added that Carbon surpasses competitors, including GPT-6 Astra and Claude Opus 5.5, across multiple technical benchmarks. [1][2][3]

Mashable, SiliconANGLE AI, and The Verge all reported that Gemini 4 Carbon secured the top spot on the Artificial Analysis Intelligence Index. Google DeepMind originally developed the Gemini multimodal large language model family as the successor to LaMDA and PaLM 2. [1][2][3][13]