Yesterday, I showed you how DeepSeek’s AI model completed the same research task as two competing AI models for a tiny fraction of their cost.
That was just one experiment, but it points to a much bigger trend taking place across artificial intelligence.
AI models aren’t just getting smarter. In many cases, they’re also getting dramatically cheaper to use.
And this week’s chart shows just how large the difference has become.
The Price of Intelligence
This week’s chart comes from Artificial Analysis, an independent research firm that tests and compares leading AI models.
As part of its Intelligence Index, Artificial Analysis runs models through the same collection of tests covering areas like reasoning, mathematics and coding.
But it also tracks something else.
How much does it cost each model to complete those tests?
Take a look.
The differences are enormous.
At the high end, Claude Opus 4.7 cost more than $5,100 to complete the Artificial Analysis benchmark suite.
Claude Sonnet 4.6 cost more than $4,200.
Gemini 3.5 Flash came in around $1,550.
But keep moving across the chart and you’ll find capable models completing the same tests for hundreds of dollars instead of thousands. That’s an enormous difference when you consider that every model is essentially being given the same assignment.
And it reinforces something that I mentioned yesterday.
The right AI model for any given task doesn’t necessarily have to be the smartest model in the world. It just has to be smart enough for the job you’re asking it to do.
Once several models can clear that bar, suddenly price matters a lot more.
I think that’s where companies are heading with AI.
They might use an expensive frontier model for difficult research or complicated coding. But answering routine customer questions, sorting documents or performing other routine tasks could call for a less powerful AI.
When you’re running those tasks millions of times, the savings can add up quickly.
That’s what makes some of the new Chinese models so interesting. It’s not like DeepSeek, Kimi and others are beating America’s best AI systems on every benchmark.
But they don’t have to.
As long as they’re good enough for the job and cost dramatically less, businesses have a powerful reason to use them.
And as I showed you yesterday, open-weight models could allow some companies to do that without sending sensitive data overseas.
But there’s another side to this story.
Here’s My Take
The cost of artificial intelligence isn’t determined solely by what a model charges. It also depends on how much intelligence that model consumes while doing the job.
That’s becoming increasingly important as AI agents take on more complicated assignments.
Because unlike a chatbot answering a single question, an agent might make hundreds or thousands of model calls before its work is finished.
In fact, new research suggests the amount of computing power an AI agent uses can vary dramatically even when it’s doing the exact same job.
So cheaper intelligence doesn’t necessarily mean smaller AI bills.
Tomorrow, I’ll show you why.
Regards,
Ian King
Chief Strategist, Banyan Hill Publishing
Editor’s Note: We’d love to hear from you!
If you want to share your thoughts or suggestions about the Daily Disruptor, or if there are any specific topics you’d like us to cover, just send an email to dailydisruptor@banyanhill.com.
Don’t worry, we won’t reveal your full name in the event we publish a response. So feel free to comment away!
