Jut to give feedback: I gave it a try on Windows 11 pc. After some adjustments (using bash over git bash for tool call, also WebSearch adjustment) and other minor patches (mainly due to JSON parsing errors because of streaming) it works fantastically!
With gpt-5 it works great. It would be a great help to mention that for using gpt-5 your organization has to be verified (https://platform.openai.com/settings/organization/general) so you do not get rate limits issue.
I am still having issues with gemini-2.5-flash and streaming (JSON parsing error, LitellmModel class does not accept the extra_body dictionary directly) but I will check with gemini to solve. Thanks for sharing.
This link helps to know the limits when using OpenAI models: https://platform.openai.com/settings/organization/limits