On Oct 29 OpenAI published Model Spec v2026.10.29, reorganizing its behavior guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. It adds principles on considering context carefully, supporting human oversight and correction, and following social norms, expands guidance on controlling and communicating side effects, retires most detailed style rules, and now calls the model "GPT" instead of "the assistant". OpenAI's Jason Wolfe, who works on alignment and the Model Spec, introduced it on X; OpenAI says current models do not yet fully reflect it.
Key Takeaways
- ✓Five values: faithfulness to human intent (the foundation), harm minimization, honesty and transparency, human agency, and care (official blog)
- ✓New "Consider context carefully" subsection: conversations, memory and personalization should be used in relevant, proportionate ways that respect privacy and user control (CHANGELOG)
- ✓"Control and communicate side effects" becomes a user-level principle with clearer guidance on authorization and trade-offs; explicit guidance on supporting human oversight and correction is added
- ✓Most detailed style guidance is retired so style adapts to user preference; stronger defaults such as approaching questions objectively remain
- ✓The subject is renamed from "the assistant" to "GPT"; OpenAI says this is target behavior that current models do not yet fully reflect
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
The Model Spec is OpenAI's public description of how its models should behave. As models move from chat to long autonomous tasks, following instructions literally is no longer enough: agents hit situations their instructions don't cover and must judge what the user really wants and which side effects are acceptable. On Oct 29 OpenAI published v2026.10.29, introduced on X by Jason Wolfe (@w01fe), whose bio says he works on alignment and the Model Spec at OpenAI.
How it works
The guidance is reorganized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. Faithfulness is the foundation and keeps the chain of command (OpenAI sets boundaries, developers direct applications within them, users direct the model within those constraints), drawing on the assistance-games framing of alignment. A new "Consider context carefully" section says memory and personalization should be used proportionately; "Control and communicate side effects" becomes a user-level principle; new defaults cover supporting human oversight and correction and following social norms. Most detailed style rules are removed, and the subject is now called "GPT".
Benchmarks
This is a behavior specification, not a model release, and OpenAI published no benchmarks. OpenAI says the Spec describes target behavior that current models do not yet fully reflect, and that results from related training methods will come in future updates.
Getting started
Agent builders should read the chain of command, side effects and authorization sections: spelling out in the system prompt which actions need confirmation is more reliable than relying on the model to guess. The full Spec is at model-spec.openai.com, the changelog is on GitHub at openai/model_spec, and OpenAI has opened a feedback form.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.