"We’ve found and resolved two issues that were affecting quality in some Claude responses. We are continuing to monitor for any ongoing quality issues.

We're grateful to the detailed community reports that helped us identify and isolate these bugs."

What were the bugs? How were they fixed?

How do we know this is even what really happened? There was a big wave of complaints on Reddit about Claude's output quality, and reports of subscription cancellations piled up.

Many people suspect some form of load optimization and/or quantized models, or other cost-optimization strategies that were applied as the cause for degradation in intelligence.

Seems like the complaints became loud enough for Anthropic to bother looking into it. But with zero transparency, zero additional info around the issue, it's hard to trust Anthropic not to continue to silently optimize for cost.

Why pay $200 a month if your "productivity boost" can randomly turn into lobotomized output overnight?