DeprecatedAI Developer Tools

Gemini 3.7 Flash (gemini-3.7-flash) and Gemini 3.5 Flash (gemini-3.5-flash)

by Google

Announced

2026-10-08

Complete

—

Refund window closes

—

Refund status

No refund applicable

Google's Gemini API changelog entry for October 8, 2026 deprecates `gemini-3.7-flash` and `gemini-3.5-flash`. Unlike a normal shutdown, neither endpoint is being turned off. Both are removed from the Google AI Studio model picker, and API requests that still name them are automatically routed to a newer model: `gemini-3.7-flash` to `gemini-3.8-flash`, and `gemini-3.5-flash` to `gemini-3.6-flash` (Google also suggests `gemini-3.8-flash` for that one). The deprecations page lists both with shutdown date N/A. Your code keeps returning 200s, but the model answering is no longer the one in your config, so behaviour, thinking options and billing follow the new model.

Refund flow

1
Not applicable. Gemini API usage is billed pay as you go per token, with no pre-paid balance tied to a model string, so there is no refund category to open.
2
Requests are billed as the model that actually serves them. On the pricing page read October 17, 2026, gemini-3.8-flash and gemini-3.6-flash share the same standard paid rates: $0.75 per 1M input tokens and $3.75 per 1M output tokens (thinking included) through December 31, 2026, then $1.50 and $7.50 starting January 1, 2027. Batch and Flex are half those rates on the same schedule.
3
Google's changelog described Gemini 3.7 Flash as launched at an introductory price through December 31, 2026, and 3.6 Flash as priced lower than 3.5 Flash. The pricing page no longer lists 3.7 Flash or 3.5 Flash, so no exact old to new delta is stated here. Compare your own invoices from before and after October 8.
4
Budget now for the January 1, 2027 step on the replacement models, since routed traffic will hit it whether or not you ever changed your model string.

Migration path

Recommended successor

Gemini 3.8 Flash (gemini-3.8-flash)

Why this is the closest fit

It is the model Google already routes gemini-3.7-flash traffic to, and the one the changelog suggests for gemini-3.5-flash users as well. Google calls it its most intelligent Flash model, generally available since September 2, 2026. It accepts the same inputs (text, image, video, audio and PDF), outputs text, and has the same 1,048,576 input and 65,536 output token limits as 3.6 Flash. Setting the string explicitly means your logs, evals and cost tracking name the model that is really running.

What differs from the original

The 3.8 Flash model page says thinking supports low, medium and high, and that minimal is not supported and returns an error. Search your code for a minimal thinking level before relying on the automatic route. Computer use is listed as Preview, and Live API, audio generation and image generation are not supported. Since the July 21, 2026 changelog, temperature, top_p and top_k are deprecated sampling parameters, so do not expect them to steer output on the newer Flash models. No quality, latency or output length comparison is claimed here, because none was tested.

Alternatives by use case

Your app named gemini-3.7-flash

Change the string to gemini-3.8-flash so config matches what is served, and remove any minimal thinking level, which 3.8 Flash rejects

Your app named gemini-3.5-flash and you want the smallest change

Pin gemini-3.6-flash, the model Google routes it to, then run your evals on it

You want the newest Flash model for agents and coding

Test gemini-3.8-flash, which Google suggests as the alternative for 3.5 Flash users too, at the same listed price as 3.6 Flash

Cost matters more than capability for high volume calls

Evaluate gemini-3.5-flash-lite, which Google positions as its low latency, cost effective subagent option

What this tool meant

Gemini 3.7 Flash was generally available for less than two months, from August 13 to October 8, 2026, before 3.8 Flash replaced it. This deprecation is quieter than a shutdown: nothing breaks on the day, which is exactly why it is easy to miss. If you track costs or quality per model name, your dashboards may still say 3.7 or 3.5 while a different model does the work.