[BUG] Significant Performance Degradation in Last 2 Weeks

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

As a power user who has processed billions of tokens over the past few months on the Pro Max tier, I've observed a sharp decline in model performance specifically within the last 2 weeks.

The Shift in Quality:
Previously, working with this model felt like collaborating with a Senior Developer - I could trust the output and focus on higher-level concerns. Now, it feels like supervising a Junior Developer where I must minutely review every single line of code, catching basic errors and unwanted additions.

Key Issues:

1. Increased iteration requirements: Tasks that previously required one attempt now require multiple iterations, even for simple implementations.

1. Unwanted code generation: The model frequently generates code I haven't requested, including:

• Additional settings that weren't specified
• Functions outside the original scope
• Features that actively contradict requirements

1. Degraded instruction following: Despite implementing elaborate planning processes (creating detailed implementation plans in separate files before execution), the model struggles to follow specifications accurately.

1. Productivity Impact: I estimate a 30-40% loss in development speed. Tasks that previously took 1-2 days now require 2-3 days due to constant corrections and re-iterations.

CLI UI Regressions:

• Missing TODO List: This was an excellent feature for tracking progress and has disappeared
• Missing token increment counter: The real-time token count on all requests was crucial feedback that processing was happening - its absence makes the interface feel unresponsive

Technical Note:
I'm aware that model weights typically aren't actively changed after deployment, which makes this degradation particularly puzzling. Whether it's context handling, system prompts, or infrastructure changes, something has fundamentally altered the user experience.

Cost-Performance Concerns:
Given the premium pricing of the Pro Max tier and the severe productivity impact, I'm actively evaluating alternatives, including Cursor, despite having invested significant time in this ecosystem.

The model's performance from just two weeks ago was excellent, making this sudden decline particularly jarring and frustrating for daily power users. I hope this feedback helps identify and address these regression issues promptly.

What Should Happen?

Please restore the model behavior and performance from 2 weeks ago - before the issues documented in:

https://www.reddit.com/r/ClaudeAI/comments/1nc4mem/update_on_recent_performance_concerns/
https://github.com/anthropics/claude-code/issues/7663
https://github.com/anthropics/claude-code/issues/6976
https://www.reddit.com/r/ClaudeAI/comments/1n29myy/is_anyone_else_experiencing_significant/
https://www.aiengineering.report/p/devs-cancel-claude-code-en-masse

This is clearly not an isolated incident but a widespread regression affecting many power users. The model from two weeks ago was exceptional - we simply need that version back.

Error Messages/Logs

Steps to Reproduce

Comparative Test: Take any moderately complex feature implementation (e.g., a REST API with validation, error handling, and database integration) that was successfully completed 3 weeks ago. Attempt to implement the same feature today with identical specifications. Observe the difference in:

Number of iterations required
Accuracy of following specifications
Amount of unwanted/hallucinated code

Frustration Metric: Analyze the percentage of conversations containing frustration indicators (all-caps messages, explicit corrections, "NO, I said..." type responses) from 3 weeks ago versus today. My usage has gone from near-zero frustration expressions to multiple daily occurrences.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

1.0.115

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_