On October 28, Google DeepMind announced Gemini 4 Carbon, arriving fourteen days after Gemini 4 Argon. Alongside improvements in general reasoning and long-horizon engineering tasks, DeepMind reported specific benchmark gains in software security. Carbon achieved 88.2 percent on SEC-bench Pro for identifying vulnerabilities, 54.8 percent on ExploitGym for autonomous exploit generation, and 71.0 percent on CWE-bench for automated patch generation.

DeepMind accompanied those capability gains with two operational choices. First, Carbon maintains the identical introductory price structure introduced with Argon: two dollars per million input tokens and ten dollars per million output tokens, with cached input tokens discounted by 95 percent. Second, DeepMind chose not to release the model directly through open API tiers.…