Which models are we talking about? Is there any degradation in quality, long context retrieval?

[flagged]