Google DeepMind released three new Gemini models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These are not flagship announcements. They are infrastructure moves, filling out the lower tiers of Google's model stack with faster, cheaper, and more specialized options.
The naming tells you the strategy. Flash means speed and cost efficiency. Lite means leaner resource demands. Cyber signals a domain-specific build, likely hardened for security-adjacent tasks. Google is not competing on one model anymore. It is competing on a portfolio, betting that different workloads need different tools.
The original post is worth reading for the specific benchmark claims and API availability details, which determine whether any of this is usable in production today or just a roadmap placeholder. The Flash-Lite positioning in particular raises questions about where it lands relative to competitors like Claude Haiku and GPT-4o Mini.
[READ ORIGINAL →]