We’ve updated the Model Spec, our public description of how we want our models to behave. The revision reorganizes its guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. It clarifies how GPT should understand human intent and exercise judgment in unfamiliar situations while remaining faithful to human direction and oversight.
Alongside clearer explanations of existing principles, the update adds guidance on context, memory, personalization, human agency, and oversight, and revises guidance on controlling and communicating side effects. It also adjusts some behavioral defaults and retires most detailed style guidance. Throughout the Spec, we now refer to the model as “GPT” rather than “the assistant,” reflecting its broader range of uses.
Read more about the changes and their rationale in ourblog post, or explore theupdated Model Spec.