Gemini 3.6 Flash

Gemini 3.6 Flash is Google's high-throughput tier, and its defining characteristic is context capacity at a price that makes using it realistic. Where other lineups make you choose between a large context window and an affordable one, Flash's proposition is both.

That changes what you build. Retrieval pipelines exist largely because context was expensive; when it is cheap, feeding the model more material directly is often simpler and more accurate than building a retrieval layer to feed it less.

Throughput and context economics. For tasks over large document sets — reviewing a folder of contracts, summarising a quarter of support tickets, extracting structure from a long report — Flash lets you put the whole thing in the prompt rather than engineering around a limit.

It is also fast enough for interactive use, which is unusual for a model with this much context capacity. Multimodal handling is well integrated rather than bolted on.

Depth of reasoning is the trade. On problems requiring several interacting inferences, Flash produces surface-level work confidently. It is the same failure mode as every fast tier: fluent and wrong is harder to catch than obviously wrong.

Large context is not the same as good use of large context. Accuracy on material buried in the middle of a very long prompt degrades, so verify rather than assume on retrieval-style tasks.

Use Flash when the task is bounded by how much material the model can see rather than how hard it has to think. It is the strongest argument in the lineup for skipping a retrieval layer entirely on small and medium corpora — worth testing before you build RAG you may not need.