22/05/2026
Microsoft, Google and Alibaba have all shipped AI models in 2026 that run on a laptop, cost nothing per use, and match ChatGPT on most business tasks.
Most UK businesses have never heard of them.
They are called Small Language Models. Microsoft’s version is called Phi-4-mini. It runs on a standard laptop and handles most routine tasks as well as the expensive models.
When you send a task to ChatGPT or Claude, that data goes to an external server. With an SLM running locally, it never leaves your building. For any UK business in a regulated sector, that is not a small detail.
The cost side works the same way. Sending GPT-5 to draft a routine email is like hiring a brain surgeon to take your blood pressure. It works. You are paying surgeon rates for a nurse’s job.
The fix is called routing. Routine tasks go to the small, fast, cheap model. Complex tasks go to the powerful one. 60 to 80% of what most businesses send to frontier AI today could move to an SLM with no noticeable quality drop.
Running everything through the expensive model is not a strategy. It is a default no one questioned.
Paying for power you do not need is still paying for it.
Follow for more AI explained simply.