Gemini 3.5 Flash is a high-efficiency multimodal large language model developed by Google, officially released on May 19, 2026. Part of the Gemini 3.5 family, it is engineered to provide frontier-level intelligence with a specific focus on speed, cost-effectiveness, and "agentic" capabilities. The minimal designation refers to a specific configuration of the model's thinking effort, a feature that allows users to modulate the depth of internal reasoning. This minimal setting is the fastest operational mode, prioritizing extremely low latency and high-volume throughput over the deep, multi-step deliberation found in higher effort configurations.
Architecturally, Gemini 3.5 Flash is built upon the Gemini 3 foundation and supports a 1,048,576 token (1M) context window. It is natively multimodal, capable of processing and reasoning across text, images, audio, video, and PDF documents within a single prompt. The model is particularly optimized for long-horizon tasks such as multi-step coding iterations and large-scale document analysis. In its minimal configuration, the model is primarily intended for chat-like interactions, rapid factual retrieval, and simple tool-calling scenarios where immediate feedback is the primary requirement.
The model introduces several advanced features, including thought preservation, which allows reasoning context to be maintained across multi-turn conversations automatically without increasing developer overhead. It supports a comprehensive suite of developer tools, including structured JSON outputs, parallel function calling, and first-party integrations such as Google Search and code execution. Despite its focus on speed, the model is designed to rival larger flagship models in coding proficiency and agentic execution, outperforming previous iterations like Gemini 3.1 Pro on specialized benchmarks such as Terminal-Bench and MCP Atlas.
Operational Guidance
When utilizing the minimal configuration, it is recommended to provide clear, direct instructions for straightforward tasks to maximize the model's speed advantages. For tasks requiring higher precision, complex mathematical proofs, or nuanced creative writing, developers can dynamically adjust the thinking level to "medium" or "high." The minimal setting is most effective for building responsive sub-agents and high-frequency loops where operational latency is the primary constraint.
The model supports a maximum output of 65,536 tokens, providing significant capacity for generating long-form content or extensive code blocks within its rapid execution framework. It is generally available through standard Google developer channels, including the Gemini API and Google AI Studio, and is integrated into enterprise-grade agent platforms.