Gemini 3.1 Flash Lite (Google Vertex AI)

Gemini 3.1 Flash Lite targets the lightweight end of Google's Gemini family, positioned as a cost- and latency-focused variant for high-frequency, agent-driven traffic. Rather than emphasizing frontier reasoning, the design centers on simple data extraction, routing, and other utility calls where response time and per-token spend matter more than raw capability. It accepts a broad mix of inputs — text, images, video, audio, and PDFs — while returning text, which makes it flexible for pipelines that need to interpret mixed documents or media snippets without committing to a larger, slower model.

In practical terms, the model pairs a one-million-token context window with a 65,536-token output ceiling, giving it room for substantial documents or long agent histories while keeping generation bounded and predictable. It supports tool calling, structured output, and reasoning steps alongside streaming, JSON mode, batch processing, and fine-tuning, which makes it straightforward to drop into existing agent frameworks that expect function calls and typed responses. The combination of a very low input price and a modest output price further reinforces its fit for high-volume workloads, where many short calls can be cheaper than relying on a larger Gemini tier without sacrificing multimodal coverage.