Sublime
An inspiration engine for ideas
🤯 Did an independent evaluation of @AnthropicAI prompt caching - and holy sh**
Built a quick "chat w/ 5 weeks of @thrusdai_pod transcripts" code to evaluate these claims & just shoved ALL those transcripts in system prompt 🤯
TL;DR - 90% cheaper sonnet & 97% cheaper haiku
Alex Volkov (Thursd/AI)x.com
Horizon Alpha from @OpenRouterAI is the new vice champion in my own Brazilian legal benchmark https://t.co/hIlZA7DFi9

It took some experimenting but I finally found a Claude 3.7 Sonnet prompt that used up most of that new 120,000 output token limit... and took 27 minutes and cost me $1.92 in API fees
One of the few times I've felt guilty about the energy usage of one of my individual prompts! https://t.co/DquvaLb8Au




