Google has launched three new artificial intelligence models, including a cybersecurity specialist that positions the company in an intensifying race over vulnerability detection. Released on July 21, Gemini 3.5 Flash Cyber is designed to detect and patch software vulnerabilities and will be offered initially to governments and trusted partners through a limited pilot. The model serves as a direct response to rival cybersecurity systems from Anthropic, OpenAI and Microsoft. During internal testing, it identified 55 confirmed vulnerabilities across Google codebases including Chrome, Android, and YouTube.
The release also includes Gemini 3.6 Flash, which improves coding and multimodal performance while using up to 17% fewer tokens and costs less per task than competing models from OpenAI and others. Gemini 3.5 Flash-Lite is the fastest and least expensive model in the 3.5 family.
The launches come as Google begins its largest-ever pretraining run for Gemini 4, while the originally planned Gemini 3.5 Pro launch has been delayed.