Howdy! I love using agent mode, especially for the test cycle. I would love to have agent support for gemini-2.5-pro-preview-03-25 using my API keys.
I was using gemini-2.5-pro-exp-03-25 but kept running into random rate limits. gemini-2.5-pro-exp-03-25 has undefined rate limits according to Google’s docs: Gemini models | Gemini API | Google AI for Developers
As far as I understand, exp and preview are the same, so it seems like adding agent support should just be a config change? I only have a cursory understanding of cursor’s architecture.
4 Likes
Gemini 2.5 Pro Exp/Preview has 5 RPM & 25 reqeusts/day limit in Free API.
Refer to Google AI Studio’s model explanation.
I have a paid API key. The issue is that cursor has a warning message saying that preview does not have agent support.
The 2.5-flash-
preview model already supports agent mode, and the 2.5-pro-
exp model—which is essentially the same as the 2.5-pro-
preview—also has agent mode support.
So, is there some significant technical barrier preventing agent mode from being added to the 2.5-pro-
preview model?
Or (and I apologize if this sounds overly suspicious) is it simply that you’re limiting features on purpose to encourage users to switch to the usage-based 2.5-pro-exp-
max mode, which offers the almost identical capabilities?
If it’s a strategic pricing decision to prioritize 2.5-pro-exp-max, I can understand that—but I’d appreciate a clear explanation of the policy.
Yes, it’s subtle, but important: 2.5-pro-exp works in agent mode, but the nearly identical 2.5-pro-preview does not. This matters a lot to us, because preview let us use our Google API billing tier for rate limits, which are much higher (enterprise grade) than what exp allows. And of course, for agent, it’s pretty easy to rack up a lot of requests.
It does seem like it should be a pretty easy config change to get preview working, and then you’re getting more of the best model into the hands of your commercial clients.
2 Likes
I have already requested support for it back when the preview version was released. I’m disappointed to see that it hasn’t been implemented yet. Also, I’d like to add that the max mode also uses the same preview model and is limited to 25 RPD when using the API key.
