The per-token price of inference is what ultimately determines how broadly an AI model gets deployed. Enterprise buyers and developers run cost models before committing to production use, and when a model's unit economics don't support a given application, that application doesn't get built. Google is pressing on that constraint with an expansion of its Gemini lineup that adds cheaper, more efficient model variants, paired with a new cybersecurity offering the company is positioning as a Mythos rival.
Cost is the argument
The move toward lower-cost Gemini models reflects a clear read of where buying decisions are currently being made. When per-query costs fall, the set of viable applications expands. Workflows that weren't worth building at a higher price point become worth building, and developers who were routing around expensive flagship models in favor of cheaper alternatives have reason to reconsider.
Google's stated goal is to close product gaps and compete on cost. Both parts of that phrase carry weight. Closing product gaps means the company has identified categories where it has not had a credible offering. Competing on cost means capability alone isn't converting buyers who have found cheaper ways to get comparable results.
More efficient models also shift the economics on both sides of the API: cheaper to serve, which compresses Google's cost of delivery, and cheaper for enterprise buyers building at scale against usage-based pricing.
The security build-out
The cybersecurity offering extends Gemini into a vertical where dedicated tooling has become the expectation. Naming it as a Mythos rival locates it in the competitive set immediately. Security operations teams have requirements that general-purpose models don't address out of the box: the tooling needs to integrate with detection and response workflows and operate within access controls that general-purpose interfaces don't enforce.
Google entering this category with a dedicated product, rather than pointing security teams at standard Gemini access, signals that the company treats a standalone security offering as table stakes for competing in that segment. The product gap it is trying to close is structural: being absent from a defined category is a different problem than being outperformed within one.