Small and large models
Situation
The situation
A small, fast model for simple tasks, a large one for hard ones.
You are the office manager of a 20-person IT services company in Lille. You use AI for many different tasks and want to know when a small model is enough.
Idea 1 of 3
Models come in different sizes
AI models are not all the same size. The number followed by “b” in their name counts their internal settings, in billions: gpt-oss-20b is a small model, gpt-oss-120b a large one. In general, the larger the model, the stronger it is, and the slower.
Idea 2 of 3
Models that reason
Some models, small or large, reason: they think in steps before answering. It helps with an analysis or a calculation. On aidock, each model’s page shows whether it “thinks before answering”: both gpt-oss-20b and gpt-oss-120b do.
Idea 3 of 3
The simple rule
Short, clear, low-stakes task: small model. Analysis, calculation, comparison or decision: large model, ideally one that reasons. If in doubt, start small and move up if the answer falls short.
Question 1 of 2
What is the advantage of a small model?
See the answer
It answers fast and is enough for simple tasks A small model is fast. It can make mistakes like a large one, especially on hard tasks.
Question 2 of 2
You are unsure which model size to use. What do you do?
See the answer
You start small, then move up if the answer falls short Starting small saves time on simple tasks. You only move up when you need to.
Exercise
Your turn
Sort each task: small, fast model or large model?
End of the lesson
Lesson complete
Simple task: small, fast model. Hard task: large model.