Google has unveiled Gemini 4 Carbon, its latest AI model, with a major claim. It has taken the top spot in a leading AI ranking. But its biggest strength may lie beyond chat, in coding and cyber defence.
Google Gemini 4 Carbon: A new leader in AI rankings
As reported by 9to5Google, Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index version 5.0, with a score of 59. The index measures how well AI models perform across different tasks. Carbon leads a list of 83 models and sits one point ahead of Anthropic’s Claude Opus 5.5, which scored 58.
The one point gap is small. The result does not mean Carbon leads every task. The launch came just two weeks after Google introduced Gemini 4 Argon. The quick release shows Google’s push to improve its AI tools.
Google Gemini 4 is built for coding and cyber defence
Carbon can handle hard coding jobs and long tasks that need several steps. These are often called agentic tasks. An AI agent can plan, use tools and check its work. The model has made gains in coding, agentic work and cyber security. Google says Carbon can work for days on complex engineering tasks. It can search large codebases, find the cause of hidden bugs and help fix them.
Cyber defence is a key part of the launch. Google says Carbon scored 88.2 per cent on SEC-bench Pro, a test for finding software flaws. It scored 54.8 per cent on ExploitGym, which checks whether a flaw can be used in an attack. On CWE-bench, which tests how well a model can fix flaws, it scored 71 per cent. Together, the tests cover finding, checking and fixing software flaws. Google also says Carbon is more than twice as successful as Argon at generating secure code in its internal tests.

Who can use Google Gemini 4 Carbon?
The model is not open to everyone yet. Google says it will first reach a small group of cyber defenders. It will then expand access to trusted business partners through its Fairwind Program. Wider access is planned, but no public date has been set. The starting price is $2 per million input tokens and $10 per million output tokens. Cached input will cost 95 per cent less. Later, rates will rise to $4 per million input tokens and $20 per million output tokens.
Google says it is also working with a voluntary US government process for testing models before release. The company says Carbon has safety tools to limit misuse, resist prompt attacks and reduce false claims. The scores should be read with care. Each benchmark checks a specific skill, and results can change as tests are updated. The first place ranking gives Google a strong start. Still, benchmark scores are only one way to judge an AI model. Users must test its speed, accuracy and value. Wider access will show how useful it is.
Author’s Opinion
Gemini 4 Carbon’s ranking is a big win for Google. However, a first place score is only the start. Its real value will depend on how well it works outside test labs. Its limited launch also means most users must wait before judging it.
READ MORE: Apple breaks Mac tradition with $2,899 MacBook Pro.
ALSO READ: Microsoft offers 60,000 employees an exit package as AI costs rise.
READ NEXT: Apple iPhone Duo buyers report cracked displays days after launch.


Comments
0Be the first to join the discussion.