Gemini 3.1 Flash Lite

Key strengths. As Google's Lite tier, it is built for speed and volume rather than peak intelligence-index scores, and independent benchmark readings for this model have varied notably across sources. Its native Google Search grounding and broad multimodal input support (text, image, audio, video, and file) make it well suited to lightweight, high-volume Inference pipelines. It also supports structured output and JSON mode, which helps when integrating it into automated pipelines that expect predictable response formats.

Limitations. Output is text-only — there's no audio, image, or video generation from this model. Its reasoning mode is optional and defaults to minimal effort, so it isn't the right pick for complex agentic, legal, medical, or research-heavy workloads where sustained multi-step reasoning matters. Training data extends to around January 2025, and Google has not disclosed parameter count or architecture details for this model.