Glossary
Benchmark
A benchmark is a standard test used to compare AI models on the same set of tasks.
What it means in practice
Scores show how models did on that test, not how they will do on your work. Treat them as a hint and run your own test.
Example
A published test of reading and maths questions used to rank language models.
Term: EvaluationGuide: Test chatbots on your own 10 tasks before you payAll terms