It has a huge context window - not sure if that can be actually used in Cursor @deanrie ? Works in agent, but not yet optimized (sonnet better).
I guess right it’s best in “ask” (or custom) to analyse complex stuff, and write to a plan.md ? Then execute with agent and sonnet 3.5
Let’s collect ideas and experiences in this thread!
Its a new model. Cursor added it. but as its new they need time to test it and set up everything so not all features are yet available.
Reading around a bit. Do I actually need an API to run this once Cursor have it optimised? It seems like it is offered as part of the cursor package. maybe I’m wrong.
Its already implemented that means you can use the model in Edit/Ask mode as it says in your screenshot. No API key or API usage directly required.
You will be able to use it in Agent once they test and adjust internally the handling for agent.
Yes part of the pacakage
They added Agent support for Gemini 2.5 Pro, but it’s a shame that image support isn’t available yet.
Yes Cursor team member mentioned in another thread that is not yet in.
Thank you for the info! 
Can we get ability use our own API key and use everything normally? Why gimp the feature when people want to use their own keys?
Supposedly they rolling it out right now, but there is no 0.48.3 yet. Not sure if he means app, or their Backend..
Also Google may still have rate limiting issues.
Is it good compared to Claude?
Too early to say.
how did you make it work, I am getting generic: “Role must be either user or assistant, but got tool”
Version? Should work from 0.48.4
The forum stripped context link + quote, TY
yup, it works now in 0.48.4 even in the Agent mode
Currently using my Gemini API key and requests are being processed through it, but the way prompts get sent through Cursor isn’t as straightforward as just typing something into chat and putting it into Google AI Studio.
Cursor needs to make sure that both the prompt and reply are sent and received in a way that seamlessly works with agent mode.
So yes, using your own API key will route requests and costs through your API key, however Cursor still needs to add agent and tool support to the model even if you’re using your API. That being said, even “un-optimized” models work in Ask mode.
Not even close. Gemini has tons of hallucinations and issues with writing code atm.
However it’s huge context window makes it incredible at generating rules and plans that can be handed off to another agent to complete the task. Until Google gets code-gen sorted out, for anything complex I use it to implement a plan in natural language that I’ll then give to claude
