Google: Gemini 3.1 Flash Lite Preview

Google: Gemini 3.1 Flash Lite Preview is a text generation model from Google, offered here as a preview build. It sits in Google's Flash Lite line, which the provider describes as its high-efficiency tier optimized for high-volume use cases. According to Google, it outperforms Gemini 2.5 Flash Lite on overall quality and approaches Gemini 2.5 Flash performance. If you are already working within the Gemini family, this model is the lightweight option aimed at workloads where you send a lot of requests and want a smaller, faster model rather than the largest one available.

The model accepts text, image, video, file and audio inputs, and it returns text only. That combination matters: you can send a prompt that includes a screenshot, a recorded clip, a document or an audio file, and get a written answer back. The context window is 1,048,576 tokens, so long documents, extended transcripts and multi-part inputs fit comfortably in a single request. The maximum output is 65,536 tokens, which is enough for long summaries, structured reports or extended reasoning traces. Supported request parameters include temperature, top_p, max_tokens, seed, response_format, structured_outputs, tools, tool_choice, reasoning, reasoning_effort and include_reasoning.

Those capabilities point to a fairly specific set of uses. High-volume classification, extraction and tagging jobs benefit from the efficiency focus, especially when the input is a mix of text and media. You can transcribe and summarize audio, describe or query video content, and parse uploaded files without a separate pipeline for each modality. Because structured outputs and response_format are supported, the model fits into systems that need JSON or schema-shaped responses rather than free prose. Tool calling makes it usable as the reasoning step in an agent loop, and the reasoning parameters let you trade off how much internal deliberation the model does per request.

You call it through api14's OpenAI-compatible API using the model id "google/gemini-3.1-flash-lite-preview". Point your existing OpenAI client at the api14 base URL, pass that id in the model field, and send a standard chat completion request with your messages. The parameters listed above are accepted, so you can set temperature and top_p for sampling, cap output with max_tokens, request deterministic runs with seed, and use tools and tool_choice for function calling. For structured responses, use response_format or structured_outputs. The reasoning, reasoning_effort and include_reasoning parameters control the model's reasoning behavior and whether that reasoning is returned to you.

One thing to keep in mind is that this is a preview release, so behavior may change before a stable version lands. It is also a Lite-tier model, which means that for the hardest tasks, such as complex multi-step reasoning or work where the absolute highest quality matters more than throughput, a larger Gemini model may be the better fit. If your workload is high volume and mostly consists of extraction, classification, summarization or media-aware processing, this model is designed for exactly that shape of work.