User reports major Claude quality drop since July 8, cause unclear
A user reports that since July 8, Claude's response quality (using the and Opus) dropped sharply and suddenly. Previously, the model could one-shot complex s with ease, but now it confuses basic principles and most of its output, a quality drop compared to what they describe as GPT-3.5-era .
Starting a fresh conversation did not help, nor did switching between shorter and more detailed prompts, nor switching to the Opus model. Tasks that had worked perfectly before failed entirely on retry.
The problem persisted the next day as well. The was used up without getting meaningful work done.
Key points
- Reports sharp quality drop in Claude ( and Opus) starting July 8
- Previously one-shot complex apps; now fails at basic principles
- New sessions, prompt changes, and switching to Opus all failed to fix it
- Problem persisted into the next day
- Burned through with no working output