Hi @Yingrong, Version Affected:
Current Version: 06/05 (June 5)
Previous Version: 05/06 (May 6) Improvement:
The 06/05 version shows significant improvement over the 05/06 version. Issue:
During longer chats, Gemini 2.5 frequently stops “thinking” (i.e., stops generating thoughtful or context-aware responses).
This occurs even when thinking and budget are enabled.
Result: Quality of responses declines sharply once the model stops thinking.
Key Quote:
“If Gemini 2.5 stops thinking, the quality will decline a lot.”