2 ms·
My take from this is that 85% of times we don't need a powerfull LLM like 4o. Or am I reading this wrong? :)
by daghamm 2y ago
My take from this is that 85% of times we don't need a powerfull LLM like 4o.
Or am I reading this wrong? :)
- thomashop 2y agoYou're reading it right. They have developed a system that automatically decides which model is sufficient, depending on your inputs, saving you costs even within one conversation stream. The OpenAI-compatible API allows you to talk to the router like a regular GPT model.