On October 28, 2026, Google announced its Gemini 4 Carbon model, but its initial launch is not open to everyone: access is first aimed at selected cyber defenders through its Fairwind program, ahead of wider availability. The decision contrasts with the usual mass launches and puts security at the center of the strategy.
According to the source, the model scores 88.2% on SEC-bench Pro, a benchmark related to finding vulnerabilities. The introductory API price reported by the source is $2 per million input tokens and $10 per million output tokens.
The approach makes practical sense: a model strong at security tasks can help defend systems, but it also demands controls over who uses it first and for what. That is why this launch is best read less as a consumer product and more as a specialized tool arriving in phases.
Source: AI Impact Hub (https://www.aiimpacthub.com/ai-news/2026-10-30)
