I have been using the gemini-1.5-flash model, and starting from November 16th, with the same prompts, the output has changed. Based on my testing, the previous output was equivalent to gemini-1.5-flash-001, and the current output is equivalent to gemini-1.5-flash-002. Is there any official documentation regarding the model switch time?
Also, gemini-1.5-flash-002 is not very stable; sometimes the delay is quite high.
The only change similar is this one in the release notes, when it says on September 24, 2024:
- Updated […] the
gemini-1.5-flash-latestmodel code to usegemini-1.5-flash-002.
But that change should be for the gemini-1.5-flash-latest alias, not for gemini-1.5-flash.
In fact, in this page it still says that gemini-1.5-flash references gemini-1.5-flash-001:
There’s definitely been a silent model change in the background from 001 to 002.
And unfortunately Google Gemini Flash 002 is dumber than 001.