Newer AI models cost more per token but often still cheaper to use

Newer AI models cost more per token but often still cheaper to use.

Why?

Because smarter models often need fewer tokens to get the job done.

Less unnecessary context. 
Shorter reasoning paths. 
Fewer retries and corrections. 
Fewer failed attempts.

A model that costs 4x more per token but solves the task using 1/10th the tokens is actually cheaper.

We keep comparing models on $/million tokens.

The metric that actually matters is cost per successful task.

Discuss on LinkedIn