User reports Claude Code got noticeably worse about a week ago
A user who relies on Claude Code for work reports that its quality dropped sharply about a week ago. Previously it handled many datasets without losing context, kept track of long conversations accurately, grounded its answers in the actual rather than guessing, gave clear and direct explanations, and reliably carried out multi-step analysis with minimal steering. Now it forgets things discussed only a minute or two earlier, speculates or assumes instead of reading the relevant files, answers from general prior knowledge rather than the project's actual context, and starts doing unrequested work instead of the task asked for.
The user has to repeatedly redirect it back to files or earlier context it used to track automatically. They note these are the exact same problems they previously had with ChatGPT before switching to Claude. Nothing about their workflow or prompts changed, the shift was sudden rather than gradual, and it started around a week ago.
They are asking whether this reflects a model update and whether others are seeing the same thing.
Key points
- User reports a sharp drop in Claude Code's data-analysis performance starting about a week ago
- Symptoms: forgetting recent discussion, guessing instead of reading files, relying on prior knowledge over , doing unrequested work
- No change to workflow or prompts; the shift was sudden, not gradual
- User says these are the same problems they previously experienced with ChatGPT
- Post asks whether this is a model update and whether others are seeing it too