What is Benchmark?
Also called: Evaluation
A benchmark is a standardised test for comparing AI models, for example in maths, coding or general knowledge.
Example
Two models solve the same 500 maths problems and their scores are compared.
For pros
Benchmarks measure only a slice of ability and can go stale or leak into training data. For your task, your own test with real examples counts most.
Sources: crfm.stanford.edu, Wikipedia · Checked on 19 October 2026
Written by AI Compare Ninja with AI assistance and edited by us. Every explanation is our own text.
