In an Agent that does tool calling if using Gemini 2.5 pro after a long time the call to the model goes in 500 error. The same code works with: xAI grok4, OPenAI GPT-5, OpenAI GPT OSS 120B (all on OCI GenAI).

I have developed a simple test case that can be easily run:

test case

To run it simply put your Compartment ocid in the python file.

With MODEL1 (gemini pro) you see the problem. Switching to MODEL2 you get quickly a correct answer from LLM