Gemini 3 Flash Preview is a high-speed, multimodal model developed by Google, optimized for low latency and high-volume processing. Released in public preview on December 17, 2025, it is designed to deliver frontier-level performance at the efficiency and price point of the Flash series. It serves as the standard, speed-oriented version of the Gemini 3 family, distinct from specialized reasoning variants.
The model features a 1-million token context window and supports diverse multimodal inputs, including text, code, images, audio, video, and PDF documents. It is engineered for agentic workflows and interactive applications, showing significant improvements in instruction following and tool use over its predecessors.
Technically, Gemini 3 Flash introduces a Thinking Level parameter that allows developers to control the depth of the model's internal reasoning. By setting this to "minimal," the model functions as a traditional non-reasoning LLM to prioritize near-instant response times and lower costs. It also incorporates Agentic Vision, a capability that enables the model to actively investigate visual data by executing code to zoom in on or annotate specific details within an image.