Overview
- Google unveiled Gemini 4 Carbon as the next frontier model and says it builds on Gemini 4 Argon with stronger long‑horizon agentic workflows and secure code generation.
- The company reported leading scores on multiple benchmarks, citing first place on the v5.0 Artificial Analysis Intelligence Index and top results on SEC‑bench Pro (88.2%), ExploitGym (54.8%), and CWE‑bench (71.0%).
- Google said internal tests show more than a twofold uplift in secure‑code success over Argon and that partners such as the Chrome engineering team and Wiz observed higher‑quality patching and web‑security performance.
- Carbon incorporates improved alignment and prompt‑injection defenses, with Google reporting large drops in misalignment and a 0.0% success rate on Gray Swan’s Indirect Prompt Injection test in its evaluations.
- Rollout will start with a subset of vetted cyber defenders through Google’s Fairwind Program, include participation in the U.S. voluntary pre‑release testing process, and carry the same introductory token pricing as Argon.