Long-form Consistency in Model Responses — Request for Stable Instance Access

I’ve been working on long-form story creation and iterative reasoning tasks using the current Pro-tier models, and I’ve noticed something that might resonate with others here.

In multi-step conversations — especially those involving recurring characters, detailed planning, or evolving story logic — consistency has become harder to maintain.
There’s often a noticeable shift in tone, reasoning pattern, or memory anchoring between sessions (or sometimes even within a single session). It feels like the model “restarts” or changes its internal logic mid-way.

I’m not sure if this is due to routing strategies or model blending under the hood, but it makes creative and logic-sensitive workflows much harder to manage.

What I’d love is a way to:

  • Select a single model instance and keep using it across sessions
  • Retain consistent tone, reasoning approach, and silence logic
  • Avoid unexpected behavioral shifts or “template resets”

Has anyone else experienced this?

Would you support a plan tier that allows locked model access for stability, even if it comes at a specialized price?

There used to be some models in the past that felt especially coherent — it would be great to have something like that again.

Thanks — I’m genuinely interested in how others handle this, and whether OpenAI might consider a solution like this for people doing long-form creative work or design-level reasoning.

1 Like