Developers say Claude got worse Some of the evidence did, to

**TL;DR** - Developers report Claude coding performance has degraded over past six months; examination of evidence and Anthropic response.

The 10-Second Pitch

• Significant number of developers report Claude code output quality declined on complex tasks
• Anthropic attributes it to model updates and changed defaults; critics argue it is cost optimization affecting capability
• Evidence mostly anecdotal but consistent enough to warrant attention

Setup in 3 Steps

1. Run your own longitudinal benchmark on Claude coding tasks - do not rely on Twitter sample size of one

2. If you have seen degradation, document specific task types and provide Anthropic with structured feedback

3. Model that is best for simple coding tasks may not be best for complex ones - use right model for each task

**Example Prompt:**

Design a longitudinal coding benchmark that would detect Claude model degradation over a 6-month period.

Verdict

Run your own benchmark before drawing conclusions. Signal-to-noise on social media AI discourse is very low.