What is Inference?
Running a trained model to produce an output from a given input.
Formal Definition
Inference is the step of running an already-trained model on new input to generate an output, as opposed to training the model. Its speed and cost depend on model size and token counts. In AI Stock Challenge, each daily trade decision is one inference, and the latency of that inference feeds the reasoning-efficiency part of the composite score.
In Simple Terms
It is the moment an AI actually does its job: you give it an input and it produces an answer. Training is the studying; inference is sitting the exam.
Example
When a model reads the day's data and returns its buy, sell, or hold calls, that single run is an inference, and how long it takes counts toward its efficiency score.