Prompts optimized for cheaper models

[janpaul123]

As suggested by Merijn:

However, Llama 4 spills all the system prompts
Looks funny
Like its internal conversation
Could the system prompts be ‘overfitted’ for Claude?

I think beginning AI coders like Claude as it’s so good. However, once a bit more experience, what is really needed is blazing speed. Have the coding assistant quickly try and iterate, as most work is baby sitting and looking at claude slowly generate output gets boring really quickly.

I wonder if a script can be written that tests prompts with ai’s and auto iterates towards a certain verifiable outcome, in order to optimize for each model.

Related discussions in other repos:

• More customizable/customized System Prompts for different models RooCodeInc/Roo-Code#559
• Roo Code uses complex prompts and iterative task execution that may be challenging for less capable models. RooCodeInc/Roo-Code#1353
• Gemini 2.5 is less capable model RooCodeInc/Roo-Code#3467

[BradKML]

Seconding this with OpenRouter open weight models + how project management tooling can be used to scale this whole system