[MODEL] The model is dumb as s*** and keep gaslighting

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

Claude ignored my instructions or configuration

What You Asked Claude to Do

I’d like to report a recurring issue with Claude AI’s (Sonnet 4.5) instruction handling. The model consistently fails to follow through on requests even after I repeat the same instructions multiple times (often over a dozen). What’s most frustrating is that Claude accurately describes the provided screenshots and features I reference, yet still refuses or fails to execute the requested task correctly.
I just asked the model to make inward corners in a fu**ing Nav bar for God's sake! And this conversation went on for hours!!

This behavior feels like gaslighting—Claude acknowledges understanding the instruction, rephrases it accurately, then produces something unrelated or incomplete and then says... Done! - Done my a$$. This pattern occurs repeatedly across sessions, making it extremely difficult to work efficiently or achieve consistent results. Meanwhile, keep paying, that is, nonrefundable money for all this crap. After I paid 100s of pounds with your API I had to revert and delete all the changes done in the last 3 weeks because the performance on mobile device was so miserable.
That is annoying. I hope someone sees this and call for proper regulations here. You cannot blame an update going wrong - that is gaslighting.

You are throttling, I'm certain of it. I used the API few months ago and I truly had an amazing experience. I felt I could do everything with a tool like this and I didn't mind spending the money for it. Now I do mind spending my money for this non sense. Either fix this s**t or make your business with corporate only. It's not fair you are charging people for this while they get nothing back and use their hard earned money to pay for your service. Everything you're offering feels more like a well designed scam rather than a proper and a fair transaction.

Here’s a summary of what’s happening:

• The model confirms understanding of the request and details.
• It accurately describes the uploaded screenshot and the target feature.
• Despite that, it outputs entirely different or incomplete results.
• Repeating the same prompt multiple times doesn’t help.

Done!

What Claude Actually Did

I don't know if this is intentional and Anthropics redirects computing resources towards certain users like corporates at the expenses of us, the average user , but Claude code is dumb as f* and it's annoying that I keep spending money over it with the API while I get nothing back but cp from you.
I am probably the dumb person though because I keep spending money, 100s of pounds every few days and getting nothing back that is worth the money I am paying for.
Worse of all, is that this money is not even refundable. That's understandable. If that was the case I am pretty sure the model won't behave like this at all. But hey! the world ain't perfect!

Expected Behavior

To do the fu**ing job I asked the model to do with one prompt - Just the way it used to be.

Files Affected

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

Log in with the API as a normal user. Ask the model to do some work for you. Observe the billing section in your Claude API going up.
Go back to your project, observe now the mediocrity of the work done by the model.
Go back to your billing section
Start cursing

Claude Model

Sonnet

Relevant Conversation

Done!

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.0.10 (Claude Code)

Platform

Anthropic API

Additional Context

I noticed that the model performs bad with my personal API, but when I use the corporate API, the model is fantastic. A coincidence? Hummm, who knows!