opus 4.6 / sonnet 4.6 produce very low quality code as compared to Opus 4.5

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

I ask it to create implementation plans (using /superpowers:write-plan). The quality of the output is dramatically lower than Opus 4.5.

What Claude Actually Did

did not create the tasks in the right dependency order (as it used to)
Many of the tasks create empty stubbed code
some tasks contradict previous tasks

Expected Behavior

It should have ordered the tasks by dependency
code snippets should include code
should not contradict previous tasks

Files Affected

none

Permission Mode

I don't know / Not sure

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

Just use the superpowers agents (github obra:superpowers)

Claude Model

Opus

Relevant Conversation

N/A

Impact

Medium - Extra work to undo changes

Claude Code Version

2.1.71

Platform

Anthropic API

Additional Context

The general quality of the latest models (Opus 4.6 and Sonnet 4.6) when used for coding has degraded dramatically. They are so bad that I'm rolling back to Opus 4.5.

Examples:

• Insists in using camel case for the next 5 questions, even though we agree to use the API wire format.
• Given a list of tasks to implement, it will randomly decide not to do some of them
• It will create empty stubs of code, even when provided with code samples

I'm tired of reading "apologies" and "you are right"