Google’s New AI Is More Advanced: What Gemini 4 Carbon Can Do
Jakarta — Google has announced its latest artificial intelligence (AI) model, Gemini 4 Carbon, which it describes as the company’s most powerful model, with capabilities in cybersecurity and handling complex, long-term tasks.
Google Chief AI Architect Koray Kavukcuoglu said Gemini 4 Carbon brings significant improvements in programming, AI agent-based tasks and cyber defence.
Carbon ranks first in the updated Artificial Analysis Intelligence Index 5.0, with a score of 59. It also performed strongly in several cybersecurity tests, covering everything from detecting software vulnerabilities to exploiting them and fixing security flaws.
In the SEC-bench Pro test, Gemini 4 Carbon scored 88.2% for finding vulnerabilities. It also scored 54.8% on ExploitGym, which tests vulnerability exploitation, and 71% on CWE-bench for fixing security flaws.
Google said these capabilities could help developers and cybersecurity teams build safer systems. In internal testing, Carbon more than doubled the success rate of secure code generation compared with Gemini 4 Argon, a model launched two weeks earlier.
Google’s Chrome team also reported that Carbon performed more than 10 times better than frontier AI models from other laboratories in generating high-quality software patches.
Beyond cybersecurity, Google has developed Carbon to handle complex, time-consuming software engineering work. The company’s internal teams reported that the model can work for days on multi-stage tasks, including identifying root causes and fixing hidden bugs in large codebases and low-level systems.
Google also reported improvements in web development. Gemini 4 Carbon rose by 89 Elo points to fourth place in the WebDev Arena, with a score of 1,768.
Despite its enhanced cybersecurity capabilities, Google is opting for a gradual rollout. From this week, the company will make Carbon available to selected cyber-defence teams before expanding access to partners through its Fairwind programme.
Google is also participating in voluntary pre-launch model testing with the United States government. The company will continue to gather feedback and refine safeguards before making Carbon available to developers, businesses and consumers through its products.
Gemini 4 Carbon will launch at the same introductory prices as Argon: US$2 per million input tokens and US$10 per million output tokens. Google is also offering a 95% discount on cached input tokens compared with the regular input-token price.
Source: jawawa.id (https://jawawa.id/newsitem/googles-new-ai-is-more-advanced-what-gemini-4-carbon-can-do-1793674254)
Originally published by CNBC Indonesia (https://www.cnbcindonesia.com/tech/20261103082933-37-781765/ai-baru-google-makin-canggih-ini-kemampuan-gemini-carbon), translated from Indonesian