# Use a cheaper model to develop a prototype, then use an expensive one to finish your product. — retained record

> Retained reddit record from reddit:ChatGPT.

Canonical: https://highsignal.app/data/records/22efd35137a3f8c9

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ChatGPT)  published 2026-09-29 11:32 ingested 2026-09-30 02:37

[original source ↗](https://www.reddit.com/r/ChatGPT/comments/1wt7nkd/use_a_cheaper_model_to_develop_a_prototype_then/)

## retained text



```
You develop a project and create a detailed plan. You begin development, but after reaching a certain point, you realize the project doesn't feel quite how you imagined it.

You notice things that need to change here and there. Maybe the UI doesn't feel right, the workflow needs improvement, or even some of the core ideas need to be reworked.

The problem is that you may have already used a lot of tokens on an expensive model developing the original idea. Now you're going to use even more tokens reworking it. Before you realize it, you've spent a considerable amount of time and money developing a prototype, changing it, testing it again, and sometimes repeating that process several times.

If this sounds familiar, I believe it might make more sense to use a cheaper model to develop the initial prototype.

The prototype doesn't need to be perfect. It just needs to work well enough for you to interact with it and get a feel for the idea. Because the cheaper model is faster and costs less, you can experiment, change things, scrap ideas, rebuild parts of the project, and figure out what actually feels right without spending as much.

Once you have a working prototype and you're confident that the core idea, mechanics, workflow, and overall feel are what you want, then you can move to the more expensive model and develop the project properly.

I feel like we now live in a time where software development can happen faster than we can fully think through every detail required to define a project.

Because of that, instead of trying to perfectly describe everything before development begins, it may sometimes be more practical to build something quickly, interact with it, learn from it, improve the idea, and only then invest heavily in the final implementation.

In other words:

Use the cheap model to discover what you actually want.

Use the expensive model to build it right.
```



document fetched 2026-09-30 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1wt7nkd/use_a_cheaper_model_to_develop_a_prototype_then
