One of the most expensive myths in the Greek market right now has to do with AI.
It goes like this: "We need the most recent, the most powerful, the most expensive model to stay competitive."
Sounds logical. It's wrong.
And it costs businesses real money on subscriptions they don't need, for capabilities they never use.
What the data says
The Stanford AI Index is the most credible annual report on the state of AI globally. The findings of the latest edition overturn the "newer is better" narrative in a very specific way:
In 2022, the smallest model that passed a specific benchmark (MMLU, score above 60%) had 540 billion parameters.
In 2024, a model hits the same score with 3.8 billion parameters.
That's 142 times smaller. Same result.
And it doesn't stop there. According to Epoch AI, LLM inference costs are dropping between 9 and 900 times per year, depending on the task.
Translated into real business consequences: the same task that had a specific cost a year ago now costs a small fraction of that.
This isn't incremental improvement. It's an overhaul of the economic logic around AI.
Why most companies don't take advantage of this
If models are getting smaller and cheaper, why are businesses paying more?
The answer is that the market isn't designed to tell you "the small one is enough." It's designed to tell you "you need the big one."
Every week a new flagship model launches. Every week a LinkedIn post explains why it's a "game changer." Every week a tech giant's marketing budget convinces you that what you were using yesterday is now obsolete.
The reality on the books:
- The majority of business use cases don't need a frontier model. They need proper architecture.
- Proper routing between different-sized models matters more than owning the biggest one.
- Access to the "latest model" is marketing. Proper selection is strategy.
This gap — between what marketing sells and what the business actually needs — is where the money gets lost.
The real job of the specialist
There's a misconception about what an AI consultant actually does.
Many think their job is to convince you to buy more technology. To recommend the flashiest tool. To present a recent frontier model and say "this will change your life."
The real job is the opposite.
It's to save you from buying the wrong thing.
To examine what your business actually does, which tasks can be automated, and which model (or models) deliver the best result at the lowest cost.
Three questions worth answering this week if you're an owner:
- Do I know exactly which tasks my AI subscription actually covers? If you don't, you're paying blind.
- Has anyone ever checked whether a cheaper model would do the same job? Or am I just paying for the brand name?
- How much of my AI budget goes to "insurance" — owning the "best" so I don't risk anything? That cost is real, and almost always excessive.
What this means for you
The technology is being democratized. Tools are getting smaller, cheaper, more capable. The critical question is no longer "which tool should I use?"
It's: "who will help me choose right — without having a financial stake in my choice?"
If you want us to take a look together at whether your business is paying more than it needs to for AI, get in touch. The first conversation is 30 minutes and free.
Newer doesn't mean better. More expensive definitely doesn't.
Better means right for you.
Sources: Stanford AI Index Report (Stanford HAI), Epoch AI inference cost analysis.