Google DeepMind officially launched three new AI models, Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber on July 21, significantly enhancing the efficiency and practical application capabilities of medium and lightweight models.

As the core of this update, the main model Gemini 3.6 Flash has achieved significant improvements in code writing, knowledge understanding, and multimodal capabilities, while reducing token usage by up to 17%, further lowering the cost of calls. Gemini 3.5 Flash-Lite is designed to provide a highly cost-effective lightweight experience; while 3.5 Flash Cyber serves as a specialized model for cybersecurity, being made available in a limited access pilot form to government agencies and trusted partners for vulnerability identification and repair.

Google's large model Gemini

Google stated that this release aims to provide enterprise customers building large-scale AI Agents with high-reliability, low-latency infrastructure support.

Notably, the long-anticipated flagship model Gemini 3.5 Pro was not included in this update. Since its update in February, it has faced internal performance challenges, leading to a delayed release. It is currently in the partner phase testing stage, and the official team has stated they aim to launch it as soon as possible.

Facing intense competition from OpenAI's GPT-5.5/5.6 and Anthropic's Claude Opus 4.8 and Sonnet 5, Google is addressing computing power and efficiency challenges through hardware and software collaboration. It has been reported that the company is developing an efficient server chip internally with the code name "Frozen v2," and has already launched the largest Gemini 4 front-end pre-training to date, demonstrating continuous investment in AI infrastructure and next-generation large models.