Google launches new Gemini Flash model and a restricted cyber version

In short: Google released new Gemini-based AI models, including Gemini 3.6 Flash and a cybersecurity-focused version that will have limited access.

What happened

Google introduced a new set of AI models in its Gemini family. The headline release is Gemini 3.6 Flash, which Google calls its most powerful Flash model so far.

Flash models are meant to be fast and less expensive to run than larger models. Google said Gemini 3.6 Flash uses about 17% fewer “tokens” than earlier versions (tokens are small chunks of text the model reads and writes, and fewer tokens usually means lower computing cost). The New York Times said the earlier Flash models were already strong at tasks like coding and financial analysis, and this version is meant to improve on that.

Google also announced Gemini 3.5 Flash Cyber, a version tuned for cybersecurity work. Google said it was fine-tuned to spot and fix security problems in computer code, and it can do this at a lower cost than bigger, more general models. In one test on a large open-source JavaScript project, Google said the model found 55 issues, including 10 vulnerabilities that other models did not detect.

Access to Gemini 3.5 Flash Cyber will be restricted. Google said it will be available only to governments and “trusted partners,” aiming to help defenders while reducing the risk of misuse.

Why it matters

AI that can find software weaknesses can help protect companies and public services, but it can also be abused like a lockpick. Google’s limited-release approach mirrors steps taken by OpenAI and Anthropic as the competition grows to build AI tools for cybersecurity.

Source: NYTimes