What falling token prices actually do to AI product economics
- Source
- x.com
- Date
What happens when intelligence gets 100X cheaper? @shashankgoyal95 sits down with @deedydas from @MenloVentures to unpack a few truths about AI economics: - Token prices are down 40%, while usage has grown 4X. - Optimizing costs too early can be the wrong move. - The best AI products will increasingly be built across multiple models.
The conversation goes beyond cost: how fast model cycles are changing, why benchmarks only tell part of the story, and what happens when AI usage is still nowhere near its ceiling.
This is New Defaults, our event series exploring the shifts reshaping how AI gets built and used. Watch the full episode on Youtube: https://t.co/V5IKrP0d6S
Context
For an AI startup still finding a product customers want, reducing model costs can be the wrong priority, argues Deedy Das, an investor in OpenRouter, in its New Defaults interview. He distinguishes this exploratory stage from an established AI product whose model bills eat into its margin.
The OpenRouter speaker proposes a concrete sequence: start with a capable model, establish a useful task, then build tests for that task and compare models. Those tests help developers judge alternatives against what their product needs to deliver. A model's general ranking cannot settle that choice, the speaker argues in the evaluation discussion.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Checking sign-in…
Loading comments…