Google unveils Gemini 4 Carbon AI model with focus on cybersecurity
Google has announced Gemini 4 Carbon, an iteration of its previous AI model with reported improvements in coding and vulnerability detection.
Google announced the development of Gemini 4 Carbon, the latest iteration in its Gemini 4 model series, arriving two weeks after the release of Gemini 4 Argon. The company indicated that the model is currently undergoing a staged rollout, limiting immediate public access.
According to technical specifications released by Google, Carbon is designed to enhance cybersecurity workflows. The company claims the model demonstrates increased efficacy in vulnerability detection and remediation compared to its predecessor. Internal benchmarks suggest the model is twice as likely to produce code without security flaws compared to Argon.
Furthermore, Google reported that its internal teams, specifically the Chrome group, observed a tenfold increase in the generation of security patches using Carbon compared to competing models. On the WebDev Arena leaderboard, the model improved its performance by 89 points, placing it fourth in current standings.
The deployment strategy remains gradual, consistent with the company's recent approach to releasing frontier-level AI systems. While performance gains are reported, full public availability remains pending further testing.