OpenAI revises Model Spec around five core values
The updated guidelines make faithfulness to human intent the top principle and trim detailed style rules, OpenAI said on October 29, 2026.
OpenAI published a revised version of its Model Spec, the public document that sets out how its models should behave, on its official blog on October 29, 2026, according to aitimes.com.
The revision reorganizes the guidance into five core values: faithfulness to human intent, harm minimization, honesty and transparency, support for human autonomy, and consideration. Faithfulness to human intent is presented as the most important of the five.
The document also calls for models to draw on conversation history and memory to grasp what a user is actually trying to achieve, rather than responding only to the literal wording of a prompt.
At the same time, OpenAI substantially reduced the detailed style and expression rules that appeared in the previous version of the Model Spec.
OpenAI said the revised guidance is not yet fully implemented across all of its current models.
Earlier versions
The Model Spec was first published in May 2024 and has been revised several times since. Anthropic, a rival developer, sets out principles for its Claude model in a written document it calls a constitution.
Quick answers
What is the OpenAI Model Spec?
It is OpenAI's publicly published set of guidelines describing how its AI models should behave. It was first published in May 2024.
Is the revised Model Spec already in effect?
OpenAI said the revised guidance is not yet fully implemented in all of its current models.