Google announces Gemini 4 Carbon AI model with a limited release

In short: Google announced Gemini 4 Carbon, a new version of its Gemini AI model, but most people cannot use it yet.

What happened

Google has announced Gemini 4 Carbon, its latest “frontier” AI model, meaning one of its most capable models. Google says Carbon is an upgrade over Gemini 4 Argon, which it announced about two weeks earlier.

Google says Carbon is better at tasks like writing code and helping with cybersecurity work. Cybersecurity means protecting computers and online services from break-ins. Google claims Carbon is more likely than Argon to produce code without security holes, and it is better at suggesting software fixes, which are like small repairs that close a gap attackers could use.

Google also says Carbon did well on several benchmark tests, which are like standardized exams used to compare AI systems. In one test focused on blocking “prompt injection” attacks (tricking an AI with hidden instructions), Google says Carbon resisted all attempts in its evaluation.

Why it matters

Google is signaling that newer AI models may be held back longer when they are good at writing code and security related tasks. For regular people, that could mean safer software over time, but it also means announcements may not equal immediate access. If Google follows the same path as Argon, Carbon may first appear for paying subscribers and for developers using the Gemini API (a way for apps to connect to the model).

Source: Arstechnica