99% Cheaper Than a Dead Horse
OpenAI's GPT-4o mini launch comes with a benchmark so cooked it barely qualifies as math.
The price war got even more absurd after this. Every provider started picking increasingly ancient baselines to make their drops sound impressive. The marketing math never got honest.
OpenAI launched GPT-4o mini today and led with this: the cost per token has dropped 99% since text-davinci-003.
text-davinci-003.
The model that GPT-3.5-turbo deprecated. The model that, by the time most people reading this announcement had a paid OpenAI account, was already being quietly shuffled toward the exits. That's the baseline. That's the 100%.
It's not a lie, technically. If you compare anything to something old enough, the numbers get impressive. My laptop is probably 99% cheaper than a mainframe from 1974. My coffee this morning is 99% cheaper than a private audience with the Pope.
What makes this particular move beautiful is the specificity — not "older models" or "models from two years ago," but text-davinci-003 by name, with a confident period after it, like they've done you the favor of the citation.
The actual comparison you'd want is against GPT-3.5-turbo, the model that ate davinci-003's lunch and then itself became the budget option everyone defaulted to. That number is presumably less cinematic.
But "significantly cheaper than the thing we already replaced" doesn't move stock, so here we are.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.