Google DeepMind has announced Gemini 4 Carbon, its strongest AI model to date, featuring state-of-the-art cybersecurity and long-horizon capabilities. Carbon ranks #1 on the Artificial Analysis Intelligence Index at 59, showing significant gains across coding and agentic tasks. The model excels in cybersecurity defense, achieving 88.2% on SEC-bench Pro for vulnerability detection and leading in Exploit Gym and CWE-bench benchmarks.
Building on the previous Gemini 4 Argon model launched two weeks prior, Carbon demonstrates the rapid pace of AI development. Google is taking a deliberate rollout approach, initially making Carbon available to a subset of cyber defenders through the Fairwind Program before broader release. The model maintains the same pricing structure as Argon, at $2 per million input tokens and $10 per million output tokens.
Carbon is designed for sustained agentic workflows, capable of running for days to tackle complex engineering tasks. It shows improved performance in coding, reasoning, and scientific domains, with a notable +89 elo point increase on Web Dev Arena, landing at #4 with a score of 1768. The model also delivers more than 2x uplift in secure code generation and 10X better patch generation compared to previous models.
Google is actively engaging in pre-release model testing with the U.S. government while strengthening safeguards alongside capability advancements.
Source: Google DeepMind Blog · Summarized by HeadlinesBriefing