At least we can get that level of performance again 😂 I haven't looked into why models seem to 'dumb down' over time. My guess would be Anthropic's limited inference capacity.