Significant regression in Claude Code quality after recent update (Opus 4.5)

Summary

After the recent version update, I've noticed a significant decline in Claude Code's performance and quality. This has made my development workflow extremely frustrating.

Issues Observed

1. Increased Hallucinations

The model now frequently makes incorrect assumptions and generates code that doesn't match the actual requirements. It often requires multiple correction cycles to get simple tasks done correctly.

2. Decreased Intelligence / Reasoning

• Simple requirements need to be explained multiple times
• The model fails to understand context that was previously handled well
• Makes repetitive mistakes on the same type of problems
• Doesn't properly analyze the full picture before making changes

3. Opus 4.5 performs worse than previous Sonnet 4.5

Surprisingly, Opus 4.5 feels less capable than the previous Sonnet 4.5 in terms of:

• Understanding complex requirements
• Making correct decisions on first attempt
• Maintaining context throughout a conversation

4. Increased Token Usage

Token consumption has noticeably increased compared to before, likely due to:

• More back-and-forth corrections needed
• Verbose/redundant outputs
• Failing to get things right the first time

Impact

These issues combined make it very difficult to work efficiently. Tasks that previously took one or two iterations now require many more cycles of correction, significantly impacting productivity.

Environment

• Claude Code CLI
• Model: claude-opus-4-5-20250101

Expected Behavior

The model should maintain or improve upon previous quality levels, not regress.