Gemini 2.5 Pro is unable to generate anything higher than 35k tokens once time or over 600 seconds

I have asked the model to help me generate a very detailed summary of a book.
The length of the book is 440K tokens. I tried many times, and the results are:

“An internal error has occurred” always appear if the generation time over 600 seconds, and the summary is unfinished. The length of the unfinished summary is around 32 tokens, with the content of first few chapters of the book.

The name of the model is Gemini 2.5 Pro Preview 05-06. The output length setting is 65536.