Gemini 4 Carbon is now available (blog.google)

Google has announced the release of Gemini 4 Carbon, its most advanced AI model to date, designed for complex tasks in cybersecurity, software engineering, and scientific research. Carbon aims to outpace its predecessor, Gemini 4 Argon, by handling intricate, long-horizon workflows, such as multi-step coding and expansive scientific inquiry, while ensuring high accuracy at a lower cost per task. It excels in generating polished outputs across various applications, from legal and financial document synthesis to advanced visual design and full-stack development. Significantly, Gemini 4 Carbon introduces stronger capabilities in cybersecurity, demonstrating impressive performance in identifying and patching vulnerabilities, evidenced by successful internal benchmarks and collaborations with security teams. Furthermore, it enhances scientific reasoning, aiding in drug research and complex protein structure predictions, potentially accelerating innovations in these fields. The model also features improved safety measures, addressing potential misuse risks, and achieving a remarkable 0% rate of prompt injection attacks. Carbon is now available via an API for developers and enterprises, with ongoing access expansion for scientists through Google Antigravity, signaling a commitment to democratizing advanced AI tools and fostering innovation across industries.