Google DeepMind has launched Gemini 4 Carbon, its strongest frontier model built specifically for long-horizon agentic coding and cybersecurity defense. The new model ranks first on the updated v5.0 Artificial Analysis Intelligence Index with a score of 59, leading fifteen out of seventeen published benchmarks against competing systems.
Carbon is the second model in the Gemini 4 lineup, designed to handle complex engineering work that runs for days on single tasks. It achieves high marks across multiple benchmarks, including a score of 96.2 percent on GraphWalks and top performance in vulnerability finding, exploitation, and patching. Google also tuned the system for full-stack web development and multimodal understanding, though access is initially restricted to a vetted group of cyber defenders rather than a public API.
To address safety concerns, Carbon features layered defenses that lower cheating attempts, prompt-injection vulnerabilities, and high-severity misalignment alerts during live internal testing. Google is participating in U.S. government pre-release testing, with broader access planned for trusted partners through the Fairwind Program.