Google: Gemini 3.1 Pro Preview (batch)

Gemini 3.1 Pro Preview (batch) is a frontier reasoning model from Google, offered under the model id google/gemini-3.1-pro-preview:batch. It sits in the text category and is built on Google's multimodal foundation, with the batch designation indicating the variant served through this endpoint. Google positions it around enhanced software engineering performance, improved agentic reliability, and more efficient token usage across complex workflows, which tells you the intended audience is developers working on demanding, multi step tasks rather than simple chat.

The model accepts a wide range of inputs: audio, files, images, text, and video. Text is the only output modality, so any task you bring to it has to end in written output, whether that is prose, code, structured data, or tool call arguments. The context window is 1,048,576 tokens, and the maximum output is 65,536 tokens. That combination of a very large input budget with a substantial output budget makes it suitable for work where you need to supply long source material and still receive a detailed response in one pass.

Practical use cases follow directly from those capabilities. Large codebases, long technical documents, recorded meetings with audio, screen captures, and video files can all be submitted as context for analysis, summarisation, or question answering. The reasoning focus and agentic reliability make it a reasonable choice for multi step workflows where the model has to plan, call tools, and keep track of state across a long conversation. Structured outputs and tool use also support pipelines that need machine readable results rather than free text.

You call it through api14's OpenAI-compatible API using the model id google/gemini-3.1-pro-preview: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. If you already have an OpenAI client, you point it at the api14 base URL, set the model field to that id, and send requests in the usual chat completions shape. Reasoning controls (reasoning, reasoning_effort, include_reasoning) let you tune how much internal deliberation the model does, while tools and tool_choice cover function calling and structured_outputs and response_format cover constrained output.

Because this is a preview model, you should treat behaviour as subject to change and validate outputs on your own data before depending on them in production. The 65,536 token output ceiling is generous but finite, so tasks that need to emit an extremely long document in a single response may require splitting the work across calls. If your workload is purely text in and text out with a smaller context, a lighter model may be cheaper and faster, and if you need image or audio output, this model does not provide it.