Model performance degradation observed in claude-sonnet-4-6 (increased basic errors, reduced comprehension)

Since approximately 2026-03-24, I've observed a significant increase in basic errors and reduced comprehension quality from claude-sonnet-4-6 in Claude Code sessions.

Specific issues observed:

• Misreading CSV column data (attributed values to wrong column/person)
• Generic error messages that didn't reflect actual root cause
• Increased back-and-forth iterations to reach correct solutions on tasks that were previously handled in fewer steps
• Overall reduction in first-attempt accuracy on multi-step tasks

Model ID in use: claude-sonnet-4-6
Platform: macOS (darwin 23.6.0)
Approximate start of degradation: 2026-03-24