Google DeepMind has released three new Gemini models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These are not flagship releases. They are positioned as efficiency-tier and specialized variants, signaling Google's push to expand the Flash family across cost, speed, and domain-specific use cases.
The naming structure tells you something. Flash-Lite suggests a smaller, cheaper inference target. Flash Cyber implies a security or threat-detection specialization. 3.6 Flash is an incremental step above the existing 3.5 line. Google is segmenting its model portfolio the way cloud providers segment compute tiers, and the reasoning behind each model's design tradeoffs is what makes the original worth reading.
The Flash family is now Google's most active development frontier, not Gemini Ultra. If that priority shift is intentional, it reflects where real enterprise and developer demand is: fast, cheap, and fit-for-purpose models, not maximal capability. Read the original for the specific benchmark numbers and deployment contexts that define each model's actual use case.
[READ ORIGINAL →]