Google Unveils Gemini 4 Carbon, Its Newest Frontier Model for Coding and Cybersecurity

Google announced Gemini 4 Carbon, its latest frontier model, following the release of Gemini 4 Argon just two weeks earlier. Carbon claims the top spot on Artificial Analysis's Intelligence Index v5.0 with a score of 59. Google attributes the rapid release cadence to its researchers and models working together to advance capabilities faster.

Carbon brings improvements across coding, agentic tasks, scientific work, full-stack web development and multimodal understanding. On web development, it jumped 89 Elo points to land fourth overall, scoring 1,769 on WebDev Arena. In cybersecurity, Carbon leads SEC-bench Pro at 88.2 percent for vulnerability detection, ExploitGym at 54.8 percent for exploitation, and CWE-bench at 71.0 percent for patching. Google also reports a strong uplift in generating secure code in live coding environments, alongside reduced misalignment and fabrication rates.

On the safety front, Carbon cuts unsupported claims by approximately 2x compared to Argon. Instead of hedging or fabricating around missing evidence, it places methodological caveats directly into outputs or clearly lists what remains required. It also reduces overreach by 26 percent and under-service by 29 percent, making multi-step task execution more reliable. When facing ambiguity, Carbon surfaces relevant discrepancies to the user before acting rather than guessing.

Cyber defenders will receive access first this week, with broader availability for developers, businesses and consumers described as "coming soon." Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 respectively at standard rates, with a 95 percent discount on cached inputs.