Definition
Model version records which underlying model or agent product produced the outputs. It should be paired with evaluation date, prompt version, and settings when possible.
The specific model release or product version used during an evaluation.
Model version records which underlying model or agent product produced the outputs. It should be paired with evaluation date, prompt version, and settings when possible.
AI products change quickly. Without model version, old results can be hard to interpret or reproduce.
A vendor updates its default model, so a support benchmark should be rerun and labeled with the new version.