Google has officially introduced three new models of the Gemini Flash family, focusing on increasing efficiency and reducing operating costs. The main working version for programming tasks and multimodal scenarios is Gemini 3.6 Flash. According to the developers, it consumes seventeen percent fewer output tokens compared to the previous generation, and savings reach sixty-five percent in specialized tests. The cost of use has been reduced to one and a half dollars per million input tokens. The model also demonstrated significant performance growth in reputable benchmarks such as DeepSWE and MLE Bench, which confirms its improved capabilities for offline interaction with computer interfaces through built-in API tools.

For mass agent scenarios, the company has released Gemini 3.5 Flash-Lite, which is positioned as the fastest and most cost-effective solution in the line. The generation rate reaches three hundred and fifty tokens per second at an extremely low cost, while the model surpasses its predecessors in tests for solving engineering problems. Special attention should be paid to the specialized Gemini 3.5 Flash Cyber development, designed to identify and eliminate vulnerabilities in the software code. Due to the dual-use features, access to this tool will be limited at the initial stage and provided exclusively to government agencies and trusted partners as part of a closed pilot program.

The new models have already been integrated into the developer ecosystem through the Google AI Studio and Android Studio platforms, and are also available to corporate clients in the Gemini Enterprise environment. For mass users, updates are gradually being implemented in the mobile application and the Google Search search engine. At the same time, the company continues to test the flagship version of Gemini 3.5 Pro with key partners and announced the start of the pre-training phase for the fourth generation architecture. These software improvements are logically complemented by hardware initiatives, including the development of a dedicated Frozen v2 processor with built-in support for artificial intelligence functions.