On 8 October 2026 Google deprecated two Gemini Flash models and pointed their names to newer versions. Requests to Gemini 3.7 Flash now reach Gemini 3.8 Flash, and an early Deep Research preview agent shuts down on 23 October. If you use Flash, your answers may change slightly.
Key facts
- `gemini-3.7-flash` is deprecated and replaced by `gemini-3.8-flash`.
- `gemini-3.5-flash` is deprecated and replaced by `gemini-3.6-flash` (or 3.8 Flash).
- Requests to the old model names are routed automatically to the new ones.
- The `deep-research-pro-preview-12-2025` agent shuts down on 23 October 2026.
- Newer Deep Research previews from April 2026 are the replacements.
What happened
In its 8 October release notes, Google deprecated two Gemini Flash models. The main replacement, Gemini 3.8 Flash, is described in an earlier changelog entry as “our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows”. Gemini 3.7 Flash was replaced by Gemini 3.8 Flash, and it can no longer be selected in Google AI Studio. Gemini 3.5 Flash was replaced by Gemini 3.6 Flash, with 3.8 Flash also suggested.
Requests to the old names are routed automatically, but Google asks developers to update the model string and review the guide for the new model.
The same notes announced the end of an early Deep Research agent, `deep-research-pro-preview-12-2025`, on 23 October 2026. Developers are asked to move to `deep-research-preview-04-2026` or `deep-research-max-preview-04-2026` by changing one parameter in their requests.
Why it matters
When a model is swapped behind the scenes, apps keep working, but the answers can shift: a little longer or shorter, a different tone, a different way of formatting lists. For a developer this is a detail. For someone who relies on an AI tool for daily work, it can mean that yesterday’s perfect email template suddenly looks different.
The best protection is a prompt that asks for a fixed format. If your prompt says exactly how long the answer should be, what sections it must have and what to avoid, the change of model matters much less. Saved, structured prompts are one of the simplest ways to keep results steady over time.
How to try it
A good everyday test is rewriting. Open our free prompt Rewrite text more clearly and paste a paragraph or an email that feels confusing. Fill the brackets with who will read it and the tone you want, for example “Reader: a new client. Tone: friendly and professional.”
The prompt asks the model to keep your meaning, make the text shorter and clearer, and list what it changed. Run it in Gemini Flash, then run the same prompt in another assistant. Compare the two answers on three points: did the meaning stay the same, is the length similar, and is the list of changes honest?
If the answers are close, your prompt is doing its job. If they are very different, add one more rule, such as a maximum number of words, and try again.
Frequently asked questions
Do I need to change anything if my app uses Gemini 3.7 Flash?
Requests are routed automatically, but Google asks developers to update the model string. Calls to `gemini-3.7-flash` already reach `gemini-3.8-flash`.
What happens to the old Deep Research agent?
The `deep-research-pro-preview-12-2025` agent shuts down on 23 October 2026. Google suggests moving to the April 2026 Deep Research previews.
Why do answers change after a model update?
A new model can write in a slightly different style or length. Prompts that ask for a fixed format, length and structure keep results more consistent.
More news
Gemini 4 Argon is generally available: what to know
Google's most capable model so far, built for complex coding, agents, math and science.
Canva Craft: Canva’s first image model, built for design
A two stage model that plans the layout and the text before it draws, now open to every Canva user.
OpenAI agents can use computers, and Codex moves to the cloud
The Agents API adds computer use, while Codex runs remote tasks and reviews pull requests.