Alphabet (GOOG.US) launches three low-cost Gemini models, focusing on cybersecurity and AI inference efficiency

According to Zhihui Finance APP, on the eve of its earnings report, Alphabet (GOOGL.US, GOOG.US) subsidiary Google officially unveiled three new Gemini models on Tuesday, covering multiple fields such as cybersecurity, programming, and cost-effective inference. This move aims to showcase the latest progress of its AI product line and respond to increasingly fierce competition from Anthropic and other AI companies.

The newly released products include Gemini 3.5 Flash Cyber, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite.

Among them, Gemini 3.5 Flash Cyber is specifically developed for cybersecurity scenarios, and can be used to identify and fix software vulnerabilities. Initially, it will only be accessible to government agencies and trusted partners via limited access tests. Google stated that this model maintains professional capabilities while offering a lower per-token usage cost than larger-scale models.

Industry insiders believe that this product has the potential to help Google narrow the gap with Anthropic in the AI cybersecurity sector. Previously, Anthropic established a leading position in the automated code security protection market with its Mythos model.

Meanwhile, Google also released Gemini 3.6 Flash. The company stated that the new model further improves performance in programming, multimodal processing, and knowledge work tasks, while token consumption is reduced by up to 17% compared to the previous generation, and the per-token cost has further decreased—helping lower the deployment costs of large-scale AI applications.

Another model, Gemini 3.5 Flash-Lite, is positioned as the fastest and most cost-effective in the Gemini 3.5 series, mainly targeting high-concurrency tasks and lightweight workloads in AI agent systems.

Analysts believe that this product matrix reflects Google's attempt to make up for some delays in AI product release times through lower costs and higher efficiency.

Artificial Analysis data shows that the price of the Gemini Flash series models is already lower than similar products from Anthropic, OpenAI, and some other AI companies. Google says the most powerful new Gemini 3.6 Flash also has a lower per-task cost than OpenAI's GPT-5.6 Terra Max, Moonshot's Kimi K3, and Alibaba (BABA.US)'s Qwen 3.7 Max.

This new product launch comes as Alphabet is about to announce its quarterly earnings, while competition among AI companies continues to intensify.

Recently, demand for Moonshot's Kimi K3 has surged in the market, and the company briefly restricted new subscriptions and API access due to insufficient computational power; meanwhile, Alibaba has also announced the upcoming release of Qwen 3.8 Max, stating that its overall performance is second only to Anthropic's latest flagship model Fable 5.

The industry believes this highlights another key factor in current AI competition: building a leading model is only the first step—companies must also have sufficient computational infrastructure to support large-scale commercial deployment of their models.

In this area, Google holds some advantages with its self-developed AI chips (TPU), cloud infrastructure, and integrated hardware-software design capabilities. However, the company has also faced bottlenecks in computational resources in the past.

It is worth noting that, according to previous media reports, Google is also developing a new type of AI chip specifically for running Gemini models, aiming to improve operational efficiency by up to 10 times and further reduce AI service costs.

In a statement, a Google Cloud spokesperson said the company team continues to explore and test new technologies to provide users with higher-performing and more efficient AI services. Although not all R&D projects will eventually go into production, this continuous innovation is an important part of Google's full-stack AI strategy.

The spokesperson stated that by collaboratively designing hardware and software from the ground up, Google is able to build highly integrated and workload-optimized AI systems.

Additionally, to address market concerns about product delays, Google also disclosed more information about its product roadmap for the first time. The company stated that Gemini 3.5 Pro has already begun partner testing in preparation for a full-scale release later. At the same time, Gemini 4 has also started the largest pre-training operation to date.