[MODEL] Unbearable trash answers of all sonnet, opus model

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 request existing codebase and follow process, also provide a value suppose to be reset but actual behavior is not.

I expect claude code reveals proper reason because its suppose to be simple work order issue.

What Claude Actually Did

CC start to guess every single step.

read one line of comment, and guessing it might be an issue -> I deny it -> read next few line and point out it might be an issue. -> I deny it -> read other line with not related action and point out that's the problem. -> i deny it -> point again first comment might be wrong.

CC just guess based on comment, not a code. I know it could be hard on troubleshooting but this behavior is nonsense.

Expected Behavior

I don't care.

Files Affected

Permission Mode

Accept Edits was OFF

Can You Reproduce This?

Not sure.

Steps to Reproduce

_No response_

Claude Model

Sonnet, Opus

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

1.0.111, 20250825 all models

Platform

Anthropic API

Additional Context

All Claude Code models are producing terrible answers.
Recently, the model’s behavior has shifted heavily toward just “guessing” after glancing at the base code.
As a result, the current model has basically turned into a useless system that only reads code and blurts out “maybe it’s like this,” and because of this, failures in seq-think are becoming more and more frequent.

Previously, I had already raised a warning that Claude Code’s answers were increasingly shifting toward flattery and excessive agreement with the user. At that time, I pointed out that Claude Code is not supposed to be a toy for users who just want to “play” with AI, but rather a tool with a strong orientation toward professional use.

I am paying for the Max plan, but if model performance is not genuinely improved while keeping prices as they are, it is inevitable that users will start leaving Claude Code. And I will be the first. If you want to survive, you’d better get your act together — for your sake, and for mine.