Gemini 3.6 Flash Released — A Comparison Guide with 3.5 Flash-Lite

Google has announced a new model lineup that makes AI agents for practical work significantly more cost-effective. The lineup includes Gemini 3.6 Flash, the main engine for general use; Gemini 3.5 Flash-Lite, which boasts extreme speed and cost efficiency; and Gemini 3.5 Flash Cyber, a specialized model for security. I will provide an easy and clear comparison of their key specs and use cases so you can decide which engine best fits your business or development projects.

Smarter and More Affordable: The Main Engine, Gemini 3.6 Flash

As Google's next-generation main workhorse, gemini-3.6-flash is a new general-purpose model optimized for coding and complex agent reasoning tasks. It is now officially available for immediate use in Google AI Studio and Vertex AI, and is being rolled out to GitHub Copilot. Its knowledge cutoff date is March 2026.

This model supports a massive input context window of approximately 1.04 million tokens, allowing it to process vast amounts of data without a hitch. It features a multimodal design that recognizes text, images, video, audio, and PDFs simultaneously, and provides a generous output limit of up to 65,536 tokens, handling long code generation or lengthy text creation with ease.

In particular, powerful tools to assist with practical automation are built-in by default. These include "thinking" options that allow you to adjust the model's reasoning level, prompt caching to save costs on frequently used prompts, real-time search and maps integration, and code execution capabilities. It also includes computer use functionality in preview, enabling the creation of advanced AI agents that can directly control screens via mouse and keyboard.

The most attractive change is the significant reduction in cost. For the paid API tier, it is priced at $1.50 per 1 million input tokens and $7.50 per 1 million output tokens. Not only is this more affordable than the $9.00 output token price of the previous-generation Gemini 3.5 Flash, but it also substantially reduces the volume of output tokens consumed for certain agent tasks, effectively lowering your infrastructure operating costs.

Extreme Cost Efficiency and Ultra-High Speed: Gemini 3.5 Flash-Lite

If you are working on a project that requires processing large volumes of data in real-time or where operating costs are the highest priority, gemini-3.5-flash-lite is a very useful alternative. This model boasts an overwhelming speed of 350 output tokens per second. It is not just fast; it also offers a significant cost advantage.

For the paid API, it is priced at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. While the rates have been adjusted very slightly compared to the previous-generation 3.1 Flash-Lite, the performance improvements make it an unrivaled choice for extreme cost efficiency. It also supports a generous input context window of approximately 1.04 million tokens, making it more than capable of processing long documents at once.

In particular, if you set the thinking level option—which allows you to control the depth of reasoning—to the minimum, the model skips unnecessary reasoning processes and completes tasks with extreme speed. With the ability to read text, images, video, audio, and PDFs, it delivers superior cost efficiency when performing bulk document classification, real-time customer service chatbots, or running large numbers of secondary helper agents that serve as bridges in complex systems.

A Security Model Inaccessible to Regular Users: Gemini 3.5 Flash Cyber

Finally, Google has unveiled a specialized model shrouded in mystery that is not available to regular developers or users: the security-dedicated engine, Gemini 3.5 Flash Cyber. Unlike the previously introduced gemini-3.6-flash or gemini-3.5-flash-lite, it is not available via any public API, and its specific pricing has not been disclosed.

This specialized model is an engine that operates exclusively within Google's AI security platform, CodeMender. Currently, it is provided on a limited pilot basis only to government agencies or a select few trusted partners. Multiple security sub-agents collaborate in parallel across up to five stages to automatically identify software vulnerabilities and even create patches immediately.

In actual internal tests at Google, it identified 55 unique vulnerabilities in the V8 engine (outperforming the existing 3.5 Flash with 47 and the competitor model Claude Opus 4.6 with 36), 10 of which were not detected by any other model. In a separate field test, the Google Cloud vulnerability research team used it to discover a remote code execution vulnerability in a public API and a memory corruption bug in a production service in just 2 hours.

A Guide to Selecting the Right Model for You

Google's latest update focuses on cost optimization and mass-processing efficiency encountered when running actual services, rather than just competing on absolute performance. For agent development where complex reasoning, coding skills, and multimodal data analysis are critical, Gemini 3.6 Flash is the recommended choice. Conversely, for secondary agents handling high-speed processing of large amounts of real-time data or performing simple summaries and classifications, Gemini 3.5 Flash-Lite is far more economical.

Here are the specs of Google's new cost-effective lineup at a glance:

By appropriately combining the two public models to suit your project goals and budget, you can build smart agent workflows that yield satisfying performance while drastically cutting operating costs. Try testing the cost-effective model that best fits your service in Google AI Studio today.