Google Adds Cheaper, More Efficient Gemini AI Models Amid Competition

Alphabet's Google is expanding its Gemini artificial intelligence lineup with three new models, including one specifically designed for cybersecurity applications and two cost-efficient versions aimed at high-volume workloads.

The new additions, announced Tuesday, include Gemini 3.5 Flash Cyber, which is built to detect and fix software vulnerabilities. This specialized model will initially be available only to governments and trusted partners through a limited pilot. Google stated the model will operate at a lower price per token than larger, more advanced models, potentially helping Google compete more directly with rivals like Anthropic that have gained traction in automated code defense.

Alongside the cybersecurity offering, Google is introducing Gemini 3.6 Flash, an improved model for coding, multimodal tasks, and knowledge work, promising up to 17% reduction in token usage and lower costs per token compared to previous iterations. The Gemini 3.5 Flash-Lite model is presented as Google's fastest and least expensive option in the 3.5 family, designed for high-throughput computation and smaller tasks within larger AI systems.

The broader rollout reflects Google's strategy to leverage price and efficiency to overcome perceived delays in releasing key AI products. Data from Artificial Analysis indicates that Gemini Flash models are already priced below comparable offerings from competitors such as Anthropic, OpenAI, and Chinese AI firms. These launches precede Alphabet's upcoming earnings report and occur as Chinese companies like Moonshot AI and Alibaba continue to gain market share.

Google is also reportedly developing custom AI chips to run its Gemini models more efficiently, a move that could provide a significant advantage in managing computational costs and scaling AI services. This strategy leverages Google's integrated approach to hardware and software development, though the company has also faced its own capacity constraints.