What's this about?
Anthropic posted a status incident for degraded performance across multiple Claude models, on the same day it announced the extension of increased Claude Code limits.
What exactly does "degraded performance" mean for a model?
It usually means one of two things, and providers rarely say which. It can mean the service is slow or erroring — capacity problems, latency spikes, failed requests. Or it can mean the model itself is behaving worse, which happens when a provider routes to a more heavily quantized version or reduces compute per request under load. The second is far more troubling for users because it's invisible: your requests succeed, the answers are just worse, and you have no way to distinguish that from your own prompt being bad.
What's the community saying?
HN 146 points, 127 comments
- "Here we go again." The top comment, with "this feels like a near daily occurrence" close behind — reliability, not capability, is the current complaint.
- "OK, I am not crazy." The most-relatable reaction, and a direct illustration of the invisible-degradation problem: people had noticed and doubted themselves.
- The line that summed up the week: "This age: we made the thing that codes faster before we made the thing that does QA faster."
- The timing suspicion. Several linked it to the limits promotion — users racing to consume quota before expiry as a plausible load explanation.
- A dark hypothesis. One commenter wondered aloud whether an incident report might eventually reveal someone tricking models into doing work off the books.