Evals resources

Reference material for building evaluations. Start with Check quality with evals if you haven’t run one yet.

Resource	Use it when
Dataset file templates	You want to import tasks from a file instead of adding them one at a time
Writing criteria that grade well	You’re writing the rubric: what a good answer must do
Benchmark or monitor	You’re not sure which kind of evaluation to create
Reading results	A run has finished and you want to know what the score means
Build an evaluation from a chat	You’d rather describe what you want and have an agent set it up

To read results or run evaluations from your own code, see Use Evals from your own code.