> While per-token inference costs for fine-tuned GPT-3.5 is 10x more expensive than GPT-3.5 it is still 10x cheaper than GPT-4!

Not quite accurate; finetuned 3.5 is only 4x cheaper than GPT-4. Cost per million output tokens from https://openai.com/pricing

  $ 2 - GPT-3.5 
  $16 - finetuned GPT-3.5
  $60 - GPT 4