Feature Request: Automatic Model & Compute Routing
With the introduction of more capable agentic models and ChatGPT Work, users now have to decide which model/reasoning level to use before knowing how complex a task will actually be.
I would like to see an optional “Auto – Optimize Model Usage” setting.
When enabled, ChatGPT would analyze the task and automatically select the most appropriate model and reasoning level based on task complexity and expected compute usage.
For example:
-
Simple repetitive tasks → lightweight/efficient model
-
Summarization and classification → mid-tier model
-
Complex reasoning → advanced model
-
Very difficult or ambiguous tasks → highest-capability model
For longer agentic tasks, ChatGPT could dynamically route different parts of the same task to different models. For example, an email cleanup task might use a lightweight model to classify hundreds of routine emails, while escalating only ambiguous or complex emails to a stronger model.
Users could choose a preference such as:
Auto – Conserve Usage
Use the least expensive model likely to complete the task successfully.
Auto – Balanced
Balance capability, reliability, and usage.
Auto – Maximum Quality
Prefer stronger models when they could materially improve the result.
Manual
User selects the model and reasoning level.
The UI could show which model was automatically selected and allow the user to override it at any time.
This would be especially valuable for ChatGPT Work. Currently, long-running tasks can consume usage allowances very quickly when a powerful model is selected, even though many individual steps do not require that level of intelligence.
The goal isn’t simply to increase usage limits. It is to use the available compute budget more intelligently.
As ChatGPT becomes more agentic, I believe model selection should increasingly become something the system can optimize for the user rather than requiring the user to micromanage the model before every task.