What Is Model Evaluation?

Model Evaluation — the systematic testing of an AI model's capabilities, limitations, and risks, using benchmarks, structured tests, and adversarial probing.

Model evaluation (often shortened to "evals") is how teams find out what a model can and cannot do before relying on it — covering accuracy on benchmarks, behaviour on edge cases, safety properties, and resistance to misuse. As frontier models have grown more capable, structured evaluation has become central to both developer safety frameworks and deployer due diligence.

Source: NIST AI RMF; frontier model safety frameworks