Google: Gemini 3.1 Flash Lite (batch)

Google: Gemini 3.1 Flash Lite (batch) is a text model from Google. It belongs to the Gemini 3.1 family, specifically the Flash Lite tier, which Google describes as a generally available, high-efficiency multimodal model built for low-latency, high-volume workloads. The batch variant is the version exposed on api14 under the model id google/gemini-3.1-flash-lite:batch, and it is aimed at workloads where throughput and efficiency matter more than running each request interactively.

The model accepts text, image, video, file, and audio inputs, and it produces text output only. Its context window is 1,048,576 tokens, which is large enough to hold long documents, transcripts, or mixed media sets alongside the instructions for a task. The maximum output is 65,536 tokens, so responses can be substantial when a task calls for long-form generation rather than a short answer. Because input modalities include video and audio as well as PDF-style files, the model can work directly on material that would otherwise need a separate transcription or extraction step.

Those capabilities point to a few practical patterns. High-volume classification, extraction, and summarization over large document sets fit the model's stated focus on efficiency. Multimodal review tasks, such as scanning images or video frames for specific content, or processing audio and PDF inputs into structured text, are also within scope. The combination of a very large context window and a 65,536 token output ceiling makes it reasonable for long-context question answering, bulk content transformation, and lightweight agentic loops where the model calls tools and returns text results.

You call it through api14's OpenAI-compatible API using the model id google/gemini-3.1-flash-lite:batch. The supported request parameters are include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, and top_p. Tool calling and structured outputs are both available, so you can define functions for the model to invoke or constrain its output to a schema. Reasoning controls let you tune how much internal reasoning effort the model applies, and seed plus temperature and top_p give you control over sampling consistency.

One thing to weigh is the batch designation. If your application needs interactive, per-request latency rather than batched or high-volume processing, a non-batch variant of the same model may be the better fit. Also note that output is text only, so tasks requiring generated images, audio, or video fall outside this model's scope. Within text generation, though, the supported parameter set and the large context window cover most standard production needs.