If Claude's felt different to you lately, you aren't alone.
One examination showed that the the depth of thinking has dropped by 67% sometime in February.
This is typically what happens if an inference provider is compute starved.
They'll reduce the quality of the model to reduce strain.
It also suggests that Anthropic may be closer to pivoting away from the more generous levels of API access that the Pro and Max subscriptions offer while they prioritize more profitable enterprise customers.
This is one of many reasons that I consider local AI to be a valuable resource.
You never know when your provider is going to reduce the quality of the model you're working with and all of a sudden you find yourself wondering if you're doing something wrong, if it's the tools you're using or the provider itself.