Thanks for your tool. Have you been able to quantify the gap between your local model and chatgpt in terms of ‘unminification performance’?
For small scripts I've found the output to be very similar between small local models and GPT-4o (judging by a human eye).