Google launches Gemini 4 Carbon in the API and the Gemini app

Google has unveiled Gemini 4 Carbon, the most advanced version of its AI to date, according to the company. Until now it was only available through the Fairwind Programme and to cybersecurity experts. It can now be used through the developer interface Google offers in AI Studio and, according to a second outlet, it also reaches businesses, scientists and general users.

According to Google, Carbon evolves from Argon and is more powerful. The company maintains that Argon remains the best choice for most tasks, especially in software engineering and knowledge work. Carbon is designed for hard problems, such as writing and debugging complex code. Published results rank it first in CursorBench 4.0, WANDR and AutomationBench, with a cost per task lower than that of the second-best model.

The launch price is the same as for Gemini 4 Argon: 2 dollars per million input tokens and 10 dollars per million output tokens. When the promotion ends, it will rise to 4 and 20 dollars. Users with the Google AI Ultra plan will be able to use it in Google Antigravity, in the Gemini app and in AI Mode in Search. On cybersecurity, Google says it is the most capable model so far. It leads the vulnerability localisation (SEC-bench Pro), exploitation (ExploitGym) and correction (CWE-bench v1) tests, and it can generate verified patches for the root cause.

The sources do not fully agree on scope: one refers to general users, while the product breakdown mentions the Ultra plan. It is also unclear how long the promotion will last or how the benchmarks were measured, since the data comes from Google and other companies cited.