Google: Gemini 3.1 Pro Preview

Google: Gemini 3.1 Pro Preview is a frontier reasoning model from Google, offered as a preview release. It sits in the text category and is built on Google's multimodal foundation, with the provider describing it as delivering enhanced software engineering performance, improved agentic reliability, and more efficient token usage across complex workflows. The provider release date is 2026-02-19.

The model accepts audio, file, image, text, and video as inputs, and it produces text as output. Its context window is 1,048,576 tokens, and the maximum output is 65,536 tokens. Reasoning is part of the model's behaviour rather than a separate mode, and the supported parameters include reasoning, reasoning_effort, and include_reasoning, which let you control or surface that reasoning in your requests.

Those capabilities point to a few practical patterns. Long documents, recordings, images, and video can be sent together in a single request, so tasks like summarising a recorded meeting alongside its slide deck, extracting structured data from scanned files, or answering questions over a large codebase fit the input modalities and the context window. Because the model is positioned around software engineering and agentic reliability, it also suits coding assistants, multi-step tool-driven workflows, and pipelines where a model has to call tools repeatedly and stay on task. Structured outputs and response_format make it a reasonable choice when downstream systems need predictable, parseable text rather than free-form prose.

You call it through api14's OpenAI-compatible API using the model id "google/gemini-3.1-pro-preview". 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. Tools and tool_choice cover function calling, response_format and structured_outputs cover constrained output shapes, and seed, temperature, top_p, and stop give you control over sampling and where generation ends.

Keep the output ceiling in mind: 65,536 tokens is the maximum output, so very large single responses need to be split or streamed across calls. The model is a preview release, which is worth noting if you depend on long term stability. If your task is plain text generation without audio, image, video, or file inputs, a simpler text-only model may be sufficient, and if you do not need reasoning controls or tool use, those parameters add nothing to your integration.