Gemini 2.5 Pro output inconsistency — Is there a frozen or stable version available?

Gemini API
1

Hi everyone,

I’ve been using the publicly accessible Gemini 2.5 Pro model via the API for a specific workflow where consistent output is important.

Recently, I’ve noticed that:

  • The same prompt,

  • The same file,

  • With no parameter changes,

…now produces different results compared to earlier. This is causing issues because my prompts were tuned for very specific, repeatable behavior.

My questions for the community:

  1. Is there a way to request a “frozen” or pinned version of Gemini 2.5 Pro so that model updates don’t silently affect output?

  2. Has anyone else experienced prompt instability or output drift even though nothing changed on your side?

  3. How are you handling consistency when the public model version evolves over time?

Any guidance or shared experiences would be really helpful.
Thanks!

3

Hi @Manav_Keswani
Welcome to Google AI Forum!!!

Thank you for bringing this to our attention.
If possible, could you please share the same prompt and file, so that we can review them on our end? Because i tested using Gemini 2.5 Pro with the some prompt and received an accurate response based on what we provided.