Google has unveiled Gemini 4 Carbon just two weeks after launching its predecessor, Gemini 4 Argon, pushing the pace of its frontier AI development. The new model features state-of-the-art capabilities in finding real-world software vulnerabilities, validating exploits, and effectively patching security flaws. Carbon claims the top position on v5.0 of the Artificial Analysis Intelligence Index and outperforms multiple competing industry systems across key technical benchmarks. Alongside its performance gains, Google highlights significant reductions in model hallucinations, overreach, and prompt-injection vulnerabilities, recording a 0% success rate on Gray Swan’s attack benchmark. Carbon is currently launching through a measured rollout first to trusted cybersecurity defenders and through the Fairwind Program ahead of wider availability.

Full report : Google’s newest flagship frontier AI model, Carbon delivers advanced long-horizon execution and cybersecurity capabiliteis.