Will you optimize the Agent to gemini-2.5-pro?
12 Likes
Please, we need an answer. Will the developers optimize cursor for agent support? Currently, cursor performance is very poor for agentic use with Gemini. Please, this is a request: don’t focus your product on just one AI (i.e., Claude Sonnet). If in the future they increase their API price from $15 to something similar to what OpenAI is doing with their recent version, you will lose all your customers and face significant financial losses as well.
8 Likes
5 models for agent so far.
If in the future they increase their API price from $15 to something similar to what OpenAI is doing with their recent version, you will lose all your customers and face significant financial losses as well.
Seems like they are following "Open"AI’s path. Cursor isn’t even open-source, so none of us can fix it.
2 Likes
This is just a brand new model. They need time to do things.
Why the negativity?
3 Likes
that brand new model better running on cline with openrouter than cursor in my opinion
2 Likes
This is also the first time google took the lead, they are the company with the most resources. They will probably be in the lead for the future.
I think it’s a global problem.
no they won’t. i will be paying lol
Exacty thats what i am trying to convey to cursor team . Its not google gemini problem , its a cursor issue that they are not optimizing it for gemini.
RooCode , Cline are working very good with 2.5 gemini .
cant use it right now for my main project runing on n8n 
I handle over 10 million tokens a day of data with over 900 executions :(((, but I think it would be an amazing model for cursor, still didnt test it for coding though
we need Image support for gemini 2.5 pro.
It supports images and many other integrations have it already with image support.
Haiku supports agentic mode too.
Maybe because they also promised V3 and R1 when they released, and how that turned out… I think it is three months in case of V3 and like two months in case of R1.
Meanwhile in Cline and Roo Code it was almost instantly. And I am pretty sure I saw an announcement surfing competition has the new gemini already added.
1 Like
Gemini 2.5 is awesome for tinkering, especially if you need long context or complex reasoning (DeepMind’s blog raves about it building simulations and games from a single prompt). But production? Nope. Indentation issues eat up time, and wonky fixes in tricky spots can break more than they solve.
The model is still raw on its own, makes dumb mistakes, and I tried using it as an agent today—it ends up breaking my code more than helping.
Hey, as others have mentioned, Gemini 2.5 Pro is a very new model, and does not behave like 2.0 did, so needs it’s own set of refining and tuning work behind the scenes.
Claude 3.5, 3.7 (and 3.5 Haiku) all worked very similarly to each other, meaning the Agent tuning we did on one passed over to the other.
We believe Gemini 2.5 to be a solid, capable model and we are presently working on improving the Agent to make it as stable as the Claude models!
3 Likes
can’t wait!