Google splits Gemini Flash across speed, cost and cyber defense

Google released three Gemini Flash models on July 21 with different jobs. Gemini 3.6 Flash was presented as an updated general model, Gemini 3.5 Flash-Lite as a lower-cost option, and Gemini 3.5 Flash Cyber as a specialist for finding, validating and patching vulnerabilities.

The two general models were made available through Google AI Studio and related developer surfaces, although rollout could vary by model and region. A shared launch date does not mean identical product access. Developers should check the current model IDs, locations, prices and quotas for each option before building against them.

Flash Cyber had a deliberately narrower path. Google DeepMind said it initially operated through CodeMender with limited access for government and trusted testing partners. It was not announced as an unrestricted public API for any developer, reflecting the sensitivity of a model designed to work directly with software vulnerabilities.

Google reported efficiency and benchmark improvements for the release. Those results are provider claims, not an independent finding that any Flash model is best for its category. The specialized cyber model also should not be compared with the general models on one broad score because its purpose, users and access controls are different.

The launch illustrates a portfolio strategy built around specialization. Users can choose for general quality, cost or a sensitive domain rather than wait for one universal flagship. That choice becomes useful only when the documentation makes the boundaries clear and independent testing measures performance under the same workload.