What is an AI model?
The model is the trained brain that every AI application runs on. There are dozens of them, and they differ a lot: the smartest model thinks deepest but is slower and more expensive, the fastest is almost free but misses nuance. On the right you see that difference in action. Making that choice well is a craft of its own, so we make it for you.
One task, three models
The task
Sort these ten customer emails by urgency.
Smartest modellike Claude Opus
thinking power€€€Sorted neatly, heavier than needed
All-round modellike Claude Sonnet
thinking power€€Sorted neatly
Fast modellike Claude Haiku
thinking power€Sorted neatly
For simple sorting the fast model gives the same result, at a fraction of the cost.
We make this call for every task, and adjust when better models come along.
What a model actually is
The engine under everything
An AI model is the result of months of training on enormous amounts of text. All the pattern recognition from What is AI? is stored in the model. It is the engine under every AI application you use.
Made by specialists
Models are made by specialised AI companies: Anthropic makes Claude, OpenAI makes GPT, Google makes Gemini. You do not buy a model like that; you pay for what you use, the way you pay for electricity per kilowatt hour.
Always in several sizes
Every maker has several models side by side: a top model, a mid-range one and a lightweight one. You can put the same question to each of them. The difference is in the quality of the answer, the speed and the price.
Anthropic · Claude
OpenAI · GPT
Google · Gemini
Mistral AI · Mistral
We are not tied to any single maker; for each task we pick the model that fits best.
Smart, fast or cheap
01The smartest models
€€€For heavy analysis and text where nuance counts: going through contracts, working out long reports, awkward customer questions. The most expensive per word, but on hard tasks they make the difference.
02The all-round models
€€The workhorses for most day-to-day work: writing emails, summarising, producing standard documents. Good quality at a fair price.
03The fast, cheap models
€For sorting, labelling and short answers in bulk. Not the deepest thinkers, but very fast and almost free per task.
And this field does not stand still: makers put out new versions every few months. A well-built solution can switch models without you noticing anything. Keeping up with that is our job: we follow every release and pick, for each task, what is the best choice at that moment.
What it costs
Models do not charge per question, but per token: small pieces of text of a few letters each. A page of text is roughly five hundred to a thousand tokens.
This is how a model cuts text into tokens
≈ 16 tokens
Thanks for your message, I will send the quote over today.
The price per token varies a lot: the smartest model can be tens of times more expensive than the fastest. That is why it pays to pick the right model for each task, as in the demo at the top of this page.
What it costs in your situation depends on how much you put through it. We work that out for you up front, so nothing catches you by surprise.
And how do you put this to work?
The model is the engine, the workflow is the machine around it. We pick the right engine for each step.
Which model fits your work?
That differs per task. In the free AI scan we find out where AI saves you time and which approach fits. No obligation.