Industry Odisha Bureau, Oct 29: In latest news, Google has introduced two artificial intelligence models, Gemini 4 Carbon and Genie 4, designed for cybersecurity. Software development and interactive virtual simulations. While Carbon focuses on complex coding and vulnerability detection, Genie 4 creates explorable digital environments with audio.
Announced on October 28, Gemini 4 Carbon is built to manage long running engineering tasks, identify software vulnerabilities and recommend fixes. Google reported scores of 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71% on CWE-bench. Altogether, this highlighted the model’s performance across cybersecurity related tasks.
Notably, the model will initially reach selected cybersecurity professionals through Google’s Fairwing programme. This will be followed by a gradual rollout to trusted partners. Its introductory API pricing is $2 per million input tokens and $10 per million output tokens. Google also plans to increase the rates after the introductory period.
Meanwhile, Genie 4 takes a different approach by generating interactive virtual worlds that users can modify through natural language prompts. Users can change weather conditions, adjust camera movements and add objects or characters while exploring a scene. The model also generates ambient sounds, music and dialogue alongside visuals.
The technology could support applications in gaming, filmmaking, education and robotics. This also includes simulation-based training. Google is exploring how Genie can work with its other AI models to create more consistent environments and enable complex simulated tasks.
Further, access to Genie 4 remains limited to trusted testers through a research preview. Google plans to bring it to AI Ultra subscribers through Project Genie and selected early-access partners through the Gemini API and Google AI Studio.
Meanwhile, the company acknowledged limitations in long term consistency, spatial accuracy, physical interactions and text rendering. AI-generated audio and video will carry SynthID watermarking to help identify synthetic content.
