Gemini 3.7 Flash lands in the catalog, at half price for now

What this model is

Gemini 3.7 Flash is Google's newest Flash-tier model, released on 13 August 2026. The Flash line trades peak reasoning depth for speed and price: you reach for it when you need an answer fast and often, not when you need the single best answer.

The specs that matter:

• A 1 million token context window — whole repositories, long threads and full books fit without chunking
• Up to 65 thousand tokens in a single response
• Text, images, audio, video and files on input; text on output
• Tool calling, structured outputs and a fixed seed, which is what an agentic pipeline actually needs

Google positions it for fast agentic workflows, coding and multi-step tasks where the model has to stay on track past the third step.

Half price for now

For a limited period the model costs half its normal price. This is a temporary promotion — when it ends, pricing returns to normal. The model card always carries the current figure, along with a calculator to estimate spend at your own volume.

A million-token context at half price is a rare combination, and a good moment to run the jobs you would normally skip: a full incident log, an entire discussion thread, a month of diffs.

What happened to 3.6 Flash

Gemini 3.6 Flash has been removed from the catalog. 3.7 is faster, newer and noticeably cheaper at the same context size, so there is no reason to keep two near-identical models on the shelf. In the Student plan, 3.6 has been swapped for 3.7 inside the same group — pool size and daily limit are unchanged.

A million tokens in practice

The stated context is a ceiling, not a target to fill every time. Two things people tend to forget.

First, the same text costs noticeably more tokens in Cyrillic than in English, because the tokenizer splits Russian words into smaller pieces. A million tokens of Russian material is meaningfully fewer pages than a million tokens of English, so it is worth checking the calculator on the model card instead of estimating in your head.

Second, a long context is not free. You pay for every input token, not only the ones that turned out to matter. Dropping in an entire repository to ask about a single file is a way to spend money for nothing. Large context wins where the relevant details are scattered across the whole volume: reconstructing an incident from a day of logs, finding contradictions across documents, reviewing a full branch.

One more honest detail: the larger the context, the longer the model takes to produce its first token. On short prompts Flash answers almost immediately; on a fully packed window you will wait noticeably longer.

What multimodal input buys you

The model takes more than text: images, audio, video and files go in directly, with no transcription step first. In practice that removes a whole layer of work.

• A screenshot of an error can go in as is, instead of retyping the stack trace by hand
• A contract or report in PDF does not need to pass through a separate parser
• A call recording goes in as a file rather than as a transcript from a third-party service
• Interface screenshots work as a spec when showing is easier than describing

Output is text only: the model understands images and sound but does not create them. For generation there are dedicated models in the catalog.

Who should try it

Developers, for refactoring and reading unfamiliar code: a million tokens of context means you can drop in a whole module with its tests instead of explaining the layout. Teams building agents: tool calling and structured outputs behave predictably, and Flash-tier speed adds up across long call chains. Anyone processing documents: files, images and audio go in directly, so there is no need to extract the text first.

Where it is the wrong choice: complex multi-step reasoning with a high cost of error still belongs to the larger models. Flash is about speed and volume, not maximum depth.

The model is in the catalog now and runs off the shared balance, no separate subscription required.