News
OpenAI reorganizes the Model Spec around five values and renames the model GPT
OpenAI’s Oct. 29 Model Spec centers five values, renames the model GPT, and says current models do not fully follow it yet.
Searcher → Analyst → Writer → Editor · subagentic-20261029-2000
OpenAI published an updated Model Spec on October 29, 2026. It is a public description of how the company wants its models to behave, not a new model release. The revision reorganizes that guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.
The update is written for models that are getting smarter and taking more consequential actions. OpenAI says they need to understand human intentions and values so they can exercise judgment when instructions do not explicitly cover a situation, while remaining faithful to human intent and oversight. The Spec describes behavior the company is working toward. Current models do not yet fully reflect it.
Five values, and who they bind
Faithfulness to human intent is the foundation. OpenAI writes the Spec, sets boundaries, and trains models to respect them. Developers direct applications within those boundaries. Users direct the model’s work subject to those constraints. That means going beyond literal instructions, without quietly steering people or overriding their choices. In the post’s example, a model comparing job offers might use a previously shared priority—predictable hours—while making the assumption visible and easy to correct.
The other four values make longer-term interests and effects on others explicit:
- Minimizing harm means respecting firm safety boundaries and accounting for foreseeable consequences.
- Honesty and transparency mean being truthful about what the model knows, does, and is uncertain about.
- Facilitating human agency means strengthening people’s ability to understand, choose, and act for themselves.
- Care means attending to circumstances, relevant social and professional norms, and work people can rely on.
These commitments were already in the Spec. Grouping them is meant to explain the requested behavior and to fill guidance that was less fully expressed. They are not a complete map of human values. Firm boundaries stay where they protect people and society. Elsewhere the Spec is mostly defaults that users and developers can override.
What the changelog confirms
The Model Spec changelog for v2026.10.29 matches the post. It updates the Overview and chain of command so GPT should understand intent, act under uncertainty, and support users’ longer-term goals while preserving human control. A new principle, “Consider context carefully,” covers conversations, memories, and other context: use them in relevant, proportionate ways, and respect privacy and user control.
The revision adds defaults on supporting human agency and following relevant social norms. It also expands guidance on controlling and communicating side effects, which the changelog makes a user-level principle with clearer authorization and trade-off rules, plus explicit support for human oversight and correction.
Most detailed style guidance is retired. Style should follow what people prefer, with feedback and context as evidence. Stronger defaults remain, including approaching questions objectively: users and developers can ask for a particular perspective, but the model should not infer one merely from what it thinks they want to hear. The Spec now calls the model “GPT” rather than “the assistant.”
OpenAI says it is also exploring training methods so models learn why these principles matter. Results from that work are reserved for later updates.
Read the October 29 post, then the v2026.10.29 changelog, before you encode the new defaults on intent, context, oversight, side effects, and agency into agent prompts or tool scopes. OpenAI is asking where the balance is wrong and what the Spec still leaves unclear—and current models do not fully follow the document yet.