Regression: Opus 4.5 "Shadow Downgrade" Significant reasoning quality drop & aggressive context compaction
Preflight Checklist
- [x] I have searched existing issues for similar behavior reports
- [x] This report does NOT contain sensitive information (API keys, passwords, etc.)
Type of Behavior Issue
Other unexpected behavior
What You Asked Claude to Do
Starting approximately January 12, 2026, I have observed a sharp degradation in the quality of claude code responses when using the Claude 4.5 Opus model. The model appears to be suffering from severe "laziness," refusal to follow negative constraints, and rapid context loss compared to performance levels from early January.
What Claude Actually Did
Starting approximately January 12, 2026, I have observed a sharp degradation in the quality of claude code responses when using the Claude 4.5 Opus model. The model appears to be suffering from severe "laziness," refusal to follow negative constraints, and rapid context loss compared to performance levels from early January.
Expected Behavior
Starting approximately January 12, 2026, I have observed a sharp degradation in the quality of claude code responses when using the Claude 4.5 Opus model. The model appears to be suffering from severe "laziness," refusal to follow negative constraints, and rapid context loss compared to performance levels from early January.
Files Affected
Starting approximately January 12, 2026, I have observed a sharp degradation in the quality of claude code responses when using the Claude 4.5 Opus model. The model appears to be suffering from severe "laziness," refusal to follow negative constraints, and rapid context loss compared to performance levels from early January.Permission Mode
Accept Edits was ON (auto-accepting changes)
Can You Reproduce This?
Yes, every time with the same prompt
Steps to Reproduce
Starting approximately January 12, 2026, I have observed a sharp degradation in the quality of claude code responses when using the Claude 4.5 Opus model. The model appears to be suffering from severe "laziness," refusal to follow negative constraints, and rapid context loss compared to performance levels from early January.
Claude Model
Opus
Relevant Conversation
Impact
Medium - Extra work to undo changes
Claude Code Version
2.1.19
Platform
Anthropic API
Additional Context
_No response_
9 Comments
i was using claude code (Opus 4.5) today for a simple UI issue. I found the issues myself after which claude code failed at - was some minor thing like a wrongly linked mock auth. Gave it all the browser code, instructions, file locations
it read through all files, changed a single header unrelated - ended reaaoning with fixed.
used up couple bucks worth of tokens for nothing.
I cancelled my subscription and immediately switched to anti gravity, I was waiting for a refund apparently our "glorious" company Anthropic redirected me to a bot which makes the final decisions for refund requests. Amazing job 👌 I eventually knew Anthropic was going to go this way(If something gains popularity it eventually ends up just like the other bad products.), after the users switched from GPT to Claude(Because of new "Karen" 5.2's Safety") Anthropic lowered the quality for the costs.
Same issue again. I’m done with this company—twice is enough. We expect a basic level of quality, today was and is stillreally bad.
Also experiencing this recently.
Same for me. Model feels much dummer then few weeks back.
There is some 2-3 days in the 2.1.22+ versions the Opus model is EXTREMELY degraded in quality. Im talking about simple behaviors like repeatedely working with the wrong file after 3-4 instructions to do otherwise. Its not a huge codebase understanding problem, its super simple instructions being understood like a dumb dog.
It is merely the globally known incompetence of Dario Amodei and his team of losers
Closing for now — this doesn't appear to be about Claude Code. Please open a new issue if this is still relevant.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.