Google launches AI model Gemini 4 Carbon focused on programming and cybersecurity tasks

·by Henderson·AI
Google launches AI model Gemini 4 Carbon focused on programming and cybersecurity tasks

Google recently released the AI model Gemini 4 Carbon, primarily targeting complex programming, software engineering, and cybersecurity tasks. This model is an improved version of Gemini 4 Argon, enhancing capabilities in code generation, reasoning, and agent workflows. Google stated that its internal team observed Carbon processing large codebases for several consecutive days, continuously identifying and fixing vulnerabilities despite encountering multiple failure paths.

In the updated fifth edition of the Artificial Analysis intelligence index, Gemini 4 Carbon received a score of 59, ranking first; in the organization’s programming agent index, it ranks just below Claude Opus 5.5. Carbon's score on the WebDev Arena leaderboard increased to 89, although it still lags behind competitors in certain specific programming benchmarks.

Gemini 4 Carbon excels in cybersecurity testing

In terms of cybersecurity, Carbon scored 88.2% in the SEC-bench Pro test for vulnerability discovery, 54.8% in the ExploitGym test for exploitability, and 71% in the CWE-bench vulnerability remediation test. Google also released several internal testing results. Compared to Argon, Carbon's success rate in generating secure code is more than double, with unsupported assertions reduced by nearly half; in an internal real-time programming environment, high-severity task deviation alerts decreased by nearly 80%.

Additionally, task out-of-bounds behavior decreased by 26%, and service underperformance dropped by 29%. In the assessed Gray Swan prompt injection benchmark, the attack success rate was 0%.

Currently, Carbon is available to select cybersecurity defenders through Google’s Fairwind program, with plans to expand its usage in the future. Its API's initial pricing is set at US$2 per million input tokens (approximately HK$16) and US$10 per million output tokens (approximately HK$78), which will be adjusted to US$4 (approximately HK$31) and US$20 (approximately HK$156) in the future, although Google has not announced a timeline for the price changes.

H
About the author
Henderson