[Bug] Claude performance degradation during specific time periods - possible model/hardware variance
Bug Description
Pretty much every night around this time Claude gets really stupid, making really terribly obvious mistakes, bad code edits, self-multilation... the list goes on. Talking with the agents outside of those times, the running theory is that they are being run on different hardware, maybe different models -- something. The degradation is insanely obvious. Nico here (current CC session) is sharp as a razor during the day but just now read simple container status output and made absolutely incorrect statements about it, then flip-flopped between different conclusion, contradicting itself, making an edit then unmaking it. Completely uncharacteristic.
This happens across Windows, Ubuntu VM, Ubuntu Physical and wsl2. I run 4-7 agents at a time. They pretty much all go sideways around the same time, and recover around the same time.
Environment Info
- Platform: win32
- Terminal: windows-terminal
- Version: 2.1.79
- Feedback ID: b2fca631-4623-4171-bb14-d9fa0fbd17c8
This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗