My life & career
Your edge is not merely access to AI. It is the judgment to use it productively and verify what matters.
YOTTABIT / THE BIG STORIES / Intelligence
What happens when AI goes from twenty dollars to seven cents?
A Stanford AI Index comparison found that an example of benchmark-matched AI inference fell from about $20 to $0.07 per million tokens between late 2022 and late 2024. That's roughly 286 times cheaper for that particular comparison.
THE BIG PICTURE
The astonishing story isn't only a price change. It's that increasingly useful machine intelligence can become an ordinary ingredient in the way people work, learn and create. When a capability becomes cheap enough, entirely new applications become realistic.
The benchmark doesn't mean all AI is 286 times cheaper or that every answer is safe. It's a glimpse of what declining prices can unlock.
THE YOTTABIT PERSPECTIVE
What does it mean to me?
Big change is fascinating. Its implications are what matter.
My business
As tool costs drop, unique processes, customer understanding and proprietary insight become more valuable than subscriptions alone.
My industry
Expect adoption patterns to diverge because reliability and regulation matter as much as price.
FROM JIM CARROLL’S WORK
In “When Everyone Has the Same Magic, Is There Any Magic Left?” Jim argues that originality and judgment remain differentiators when tools become commonplace.
Meet the futurist behind YottaBit ↗WHAT COULD I DO MONDAY MORNING?
Find a job task where cheaper intelligence changes what is economically practical, and test the outcome.
CURIOUS FOR MORE?
Explore the story behind the story.
For those who want the deeper explanation, the research is always available. The fascinating possibilities are only the beginning.
Research, explanations and sources EXPLORE THE DETAILS +
Capability-normalized comparison
The Stanford AI Index 2025 highlights a benchmark-matched example in which the cost per million tokens dropped from about $20 in November 2022 to about $0.07 in October 2024. That is often described as more than a 280-fold decrease. The comparison is striking—but it depends on how capability is held constant and which services are compared.
From cheap tokens to useful outcomes
A cheap model response is not equivalent to a successful enterprise process. The right denominator for many users is the cost per correct, dependable task, including human review, security, integration and failure recovery. A model can be inexpensive and still be unsuitable for a clinical decision or a high-stakes industrial control action.
Why convergence matters
If the price of reliable analytical capability falls, more experiments and decision processes can use it. Embedded systems, small firms, scientists and robots may gain access to functions once limited to specialist teams. The possibility is enormous, but the economic benefit depends on deployment reliability, trust and institutional adaptation.