Gemini 3.6 Flash: check capabilities, context and model lifecycle
A large input window does not add speech generation or Live API support. Match the exact capability table to your task.
In brief
Gemini 3.6 Flash accepts several input types but its documented output is text. Read the actual capability table instead of assuming “multimodal” means it can generate every modality.
This prevents choosing a model for speech or live interaction that the particular endpoint does not support.
Read supported output first
The model card distinguishes audio input from audio generation and Live API support. Use a supported speech or Live model when those outputs are required. Do not infer a capability from a nearby model with a similar name.
Use a large context deliberately
Supply relevant material with identifiable sections and ask for source references. Check whether the answer uses the right passage, not just whether the upload fits. Separate contradictory documents and label dates so an old statement is not treated as current.
Track the model lifecycle
Consult Google’s deprecation schedule and migration guidance before changing an existing integration. Keep the exact model string in configuration and validate supported parameters. An automatic routing change in another model is not proof that this endpoint has shut down.
Before you act
At review, Google lists 3.6 Flash as stable with no announced shutdown date, alongside newer models. A context limit is not a guarantee of perfect retrieval from every supplied page.
What to do next
- Check the exact model code and current lifecycle page.
- Match required input, output and tools with its capability table.
- Test representative inputs and output validation before deployment.
