I know the response to this will be a copy-pasted paragraph starting with “To ensure reliable performance for everyone, we are limiting access to the 2.5 models to users who have actively used them in the past.”
I’ve already read it. That exact sentence has been sent to every person raising this issue, word for word, without addressing a single specific question. I’m documenting that here so when it shows up again, there’s a clear record that it doesn’t constitute an answer and won’t be accepted as one. That response contains an internal contradiction your team refuses to resolve.
Your team claims that access is limited to active legacy users and that this restriction applies to Google AI Studio. Both statements cannot be true the way you’re presenting them.
If active users retain access, some users should still see Gemini 2.5 Pro in their model selector. They don’t. Across hundreds of replies spanning multiple days in the public thread, not one person has reported retaining UI access in Google AI Studio.
The active user criteria does not exist in practice. Either your metric captured essentially no one, or the policy as communicated is directly contradicted by the evidence shown. Both possibilities require a direct answer.
Directing users to the API endpoint is not a solution. I’m a paying Google One AI Pro subscriber. My usage has been through the AI Studio chat and build interfaces. That is the product I pay for. A raw API endpoint requires separate billing, pay-per-token costs outside my subscription, and complete workflow reconstruction. Telling a paying UI user that the API still works is deflection.
Here’s what makes this inexcusable:
Anthropic officially retired Claude Opus 3 on January 5, 2026. They publicly acknowledged that maintaining retired models is costly and operationally complex and they did it anyway. Claude Opus 3 is available to every paid Claude.ai subscriber today, transparently, without restriction, without an active user metric, and without requiring API migration.
That model is formally retired. Gemini 2.5 Pro is not deprecated by your own statement. Paying Google subscribers have less access to a non-deprecated model than Anthropic gives to a model that has been retired for eight months.
I know Gemini 4 Pro is in post-training and that RLHF, RLAIF, and safety eval require the same TPU pod architecture as inference. If Google cannot simultaneously run 2.5 Pro UI inference at scale, 3.x inference at scale, and 4 Pro post-training on fixed hardware, say it formally. Don’t hide behind vague active user restrictions that exclude everyone.
My demands are specific:
-
Publish the exact technical metric defining an active user under this restriction.
-
Confirm whether any paid AI Pro subscribers currently retain UI access to Gemini 2.5 Pro in Google AI Studio. If yes, what differentiates their accounts from those who lost access. If no, admit the active user policy is functionally a full deprecation.
-
Restore Gemini 2.5 Pro UI access for paid AI Premium subscribers until Gemini 4 Pro is available on the subscription tier if possible.
The model is running on Google’s infrastructure. Google’s team is already serving it to some users via API. Extending that to paying UI subscribers on AI Studio is the minimum standard your direct competitor already meets for a retired model. If you cannot do that, be honest about why instead of hiding behind policies and corporate language that don’t match what paying users are actually experiencing.