Gemini 3.7 Flash: API Specs, Pricing, Tools, and What Changed
Gemini 3.7 Flash is Google’s stable 1M-context workhorse with a 65,536-token output limit, configurable thinking, and broad tool support. Use it for production coding and agent workflows; validate tool behavior and pricing in your own workload before migrating every request from 3.6 Flash.
By Todd · Published 2026-08-14 · Verified 2026-08-14
| Endpoint | gemini-3.7-flash |
|---|---|
| Context | 1,048,576 input tokens |
| Max output | 65,536 tokens |
| Inputs | Text, image, video, audio, PDF |
| Thinking | Low, medium, and high levels |
| Verified | 2026-08-14 |
What changed from Gemini 3.6 Flash
The stable 3.7 endpoint keeps the 1M-token context developers expect while adding a clearer thinking control surface and a broader production-tool story. The practical change is less about a new chat persona and more about giving agent builders one stable endpoint for long inputs, structured output, and tool-assisted work.
- Stable endpoint rather than a preview route
- 1,048,576 input-token context and 65,536-token maximum output
- Low, medium, and high thinking levels
- Function calling, code execution, search grounding, file search, structured output, and preview computer use
Who should use it
Choose Gemini 3.7 Flash when a workload combines large documents or codebases with multimodal input and tools. It is a sensible first candidate for coding agents, document extraction with grounding, and workflows where one request may include text, images, audio, video, or PDFs.
- Production coding and agent loops
- Long-context multimodal analysis
- Grounded search and structured extraction
Who should not switch yet
Do not migrate every 3.6 request solely because the version number changed. Keep 3.6 for workloads with stable, already-tested prompts until you have checked output shape, tool-call frequency, latency, and the effective price basis for your traffic. Google’s model documentation establishes capabilities; it is not a substitute for a controlled regression test.
Migration checklist
Treat the migration as an endpoint and behavior change, even where the request schema is familiar.
- Pin the endpoint explicitly to gemini-3.7-flash
- Re-run JSON and function-call validation
- Test low, medium, and high thinking on representative prompts
- Measure multimodal payload handling and retry behavior
- Compare the live price owner before changing budgets
Known limits and open questions
The published model page does not establish a first-party All AI Ask benchmark verdict. We therefore do not call Gemini 3.7 Flash fastest or best here. Promotion claims, regional availability details, and workload-level latency should be checked against the current Google pricing and API documentation before procurement.
Explore the canonical data
Test it yourself
Run the same prompt across gemini 3.7 flash, gemini 3.6 flash, claude sonnet 5 in one workspace.
Open the comparison playground →FAQ
What is the Gemini 3.7 Flash context window?
The documented input context limit is 1,048,576 tokens, commonly displayed as 1M.
Does Gemini 3.7 Flash support tools?
Yes. The documented capabilities include function calling, code execution, search grounding, file search, structured output, and preview computer use.