Gemini model IDs for the Gemini API

Google calls the model ID a “model code”: the string you pass as model to the Gemini API, such as gemini-3.8-flash. This page lists the text models on the Gemini Developer API (Google AI Studio keys) from Google’s models and deprecations pages. Image, audio, music, video and embedding models are left out.

Stable, preview, latest and experimental codes

Google sorts every code into one of four version patterns:

• Stable codes point to a specific model and usually don’t change. Google says most production apps should use one, such as gemini-3.8-flash.
• Preview codes, such as gemini-3.1-pro-preview, may be used in production, typically have billing enabled, may have tighter rate limits and can be deprecated with at least two weeks’ notice.
• Latest aliases, such as gemini-flash-latest, are hot-swapped with every new release of a model variation. Google emails two weeks’ notice before a breaking change behind one.
• Experimental codes aren’t suitable for production and can disappear at any time.

Google notes this naming scheme dates from September 2025, so older models may follow different patterns: use the exact code from the list.

Which code to use

For new projects Google recommends Gemini 3.8 Flash (gemini-3.8-flash) or Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite). Both take up to 1,048,576 input tokens and write up to 65,536 output tokens. The Gemini 2.5 models are not deprecated, but Google now limits them to accounts that have used them before, so a new key may not be able to call gemini-2.5-pro.

Codes that now route to another model

Two recent codes are kept alive by redirection: requests to gemini-3.7-flash go to gemini-3.8-flash, and requests to gemini-3.5-flash go to gemini-3.6-flash. Your code keeps working, but you get a different model from the one you named, with its own behaviour. Google’s quickstart still uses gemini-3.5-flash in its examples, so copy-pasted code may already be redirected.

Shutdown dates

Google’s deprecations page lists the earliest date each model might be shut down and says the exact date is confirmed with advance notice. Gemini 3.1 Flash-Lite, for example, can shut down from 7 May 2027 and the replacement is gemini-3.5-flash-lite. Once a model is shut down its endpoint stops responding: gemini-2.0-flash was switched off on 1 Jun 2026, with gemini-3.6-flash as the replacement.

Gemini on Google Cloud

Gemini models are also offered on Google Cloud’s Agent Platform (formerly Vertex AI), which has its own catalogue and lifecycle; Gemini 3.7 Flash, for example, is listed there as generally available. This page covers the Gemini Developer API, so check Google Cloud’s model pages before using these codes there.

Try a model for free in the Gemini API playground with a free Gemini API key, see which models your key can call with the API key checker, or compare prices in the model comparison.