Google: Gemini 3 Flash Preview (batch)

Google: Gemini 3 Flash Preview (batch) is a text generation model from Google. It belongs to the Gemini 3 family and is offered in a preview and batch variant, which is the form available under this model id. The provider describes it as a high speed, high value thinking model built for agentic workflows, multi turn chat, and coding assistance, with near Pro level reasoning and tool use. It was released by the provider on 2025-12-17.

The model accepts text, image, file, audio, and video as input, and it returns text only. Its context window is 1,048,576 tokens, and the maximum output is 65,536 tokens. Because reasoning is part of the supported parameter set, the model can expose its thinking process through include_reasoning and can be tuned with reasoning and reasoning_effort. This combination of a very large context and mixed input modalities means you can feed long documents, screenshots, recordings, or clips into a single request and get a written answer back.

Those capabilities point to a few practical uses. Agentic workflows fit well, since the model supports tools and tool_choice, so it can decide when to call a function and incorporate the result. Multi turn chat is another natural fit, helped by the large context that lets a long conversation stay in one window. Coding assistance follows from the same strengths, including structured_outputs and response_format for producing machine readable results. Document and media analysis is possible too, because images, files, audio, and video can all be supplied as input alongside text.

You call it through api14's OpenAI-compatible API using the model id "google/gemini-3-flash-preview:batch". The request parameters it supports are include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, and top_p. If you are porting existing OpenAI-style code, the usual chat completions shape applies, with tools and tool_choice for function calling, response_format or structured_outputs when you need constrained output, and seed, temperature, and top_p when you want to control sampling.

Keep the output ceiling in mind: 65,536 tokens per response, even though the input context is far larger. The model produces text only, so anything requiring generated images or audio would need a different model. As a preview and batch variant, it is best treated as a candidate to evaluate against your own workload rather than a settled default, and if you need a non preview or non batch option, another model on api14 may suit better.