On 28 October Google announced Gemini 4 Carbon, a frontier model focused on cybersecurity and long-running programming tasks. At first it is available only to selected cyber defence organisations through the Fairwind programme, with broader access for developers, businesses and consumers to follow. Introductory API pricing is $2 per million input tokens and $10 per million output tokens. Prices will later double, to $4 and $20 respectively.

What it means: Giving the model to cyber defence teams first is no accident. The most capable models now find flaws in code at a speed that worries software makers, and Google is trying to give those who fix problems a head start before the tool reaches everyone. Other labs now follow the same pattern, which shows that releasing a model is increasingly treated as a security question. For Greek development teams, the practical consequence is that the time between discovering a flaw and exploiting it keeps shrinking. Regular library updates, automated dependency checks and fast patching become necessary rather than optional. On cost, the introductory price is lower than comparable rival models, but the doubling already announced means anyone building a product on it must plan for the later price. Until access opens up, comparisons with other models rest mainly on Google's own figures, which need independent confirmation. Teams that already use AI coding tools should also decide in advance which code and data they may send to an external model.

Read the original on Google

Related service: Custom software

Related stories