Training is learning. Inference is using what was learned.
Direct answer
Inference is the process of running a trained AI model on new input to get an output, such as a reply to your prompt. Training is the earlier, much heavier process of adjusting the model on data. Inference speed and cost depend on model size, input length and hardware.
Training vs inference
| Training | Inference | |
|---|---|---|
| Purpose | learn patterns | produce outputs |
| When | before release | every time you use it |
| Cost | very large, one-off | smaller each time, adds up with usage |
What affects it
- Model size: bigger is usually slower and costlier.
- Prompt and output length: more tokens mean more work.
- Hardware: specialised chips speed it up.
- Batching and caching: reduce cost.
Where Teyro fits
Teyro's The AI Stack course turns this into short lessons you can play in a few minutes a day. It has 100 lessons, and you can start free.
The bottom line
When you use a chatbot, you are using inference. Costs and speed shape what products are possible.
