Google: Gemini 3.5 Flash Lite (batch)

Gemini 3.5 Flash Lite (batch) is a text model from Google, offered here under the model id google/gemini-3.5-flash-lite:batch. Google describes it as a high-efficiency model with upgraded agentic capabilities, and positions it for subagents that execute focused tasks inside larger, multi-agent workflows. If you are building systems where one orchestrating model delegates narrow jobs to helper models, this is the kind of model Google intends for that role.

The model accepts text, image, video, file and audio as input, and produces text as output. Its context window is 1,048,576 tokens, which is large enough to hold long documents, transcripts or multi-turn histories in a single request. The maximum output is 65,536 tokens, so long generated responses are possible within that ceiling. Supported request parameters include max_tokens, stop, seed, response_format and structured_outputs, along with tools and tool_choice for function calling, reasoning, reasoning_effort and include_reasoning for control over reasoning behaviour.

Those capabilities point to a few practical patterns. Because the input side covers text plus image, video, file and audio, you can send mixed media material for summarisation, extraction or question answering rather than routing each modality through a separate pipeline. The large context window suits work over long inputs such as full codebases, lengthy reports or recorded meeting content. Tool calling and structured outputs make it a reasonable fit for subagent tasks that must return data in a fixed shape or call out to other systems, and the reasoning controls let you dial effort up or down depending on how much deliberation a task needs.

To call it through api14, use the OpenAI-compatible chat completions interface with the model id google/gemini-3.5-flash-lite:batch. Pass your inputs as messages in the usual way; the supported input modalities are declared at the model level, so text and the media types listed above are what you can supply. On the request side you can set max_tokens, stop and seed, choose a response_format or use structured_outputs when you need constrained output, and attach tools with tool_choice for function calling. Reasoning behaviour is controlled through reasoning, reasoning_effort and include_reasoning. The batch suffix in the model id indicates this variant is intended for batch style usage.

Keep the output ceiling in mind: if a task needs a single response longer than 65,536 tokens, this model will not fit and you would need one with a higher output limit. The output modality is text only, so it cannot generate images, audio or video. For work that depends on those output types, or on capabilities not represented in the parameter list above, another model may be the better choice.