This is the first benchmark that aligns with what I see in the field when the bar is set high for a task.
"frontier models fail to complete tasks with senior-level correctness and taste over 75% of the time"
@SimonPB I think you use case is a good match for this and you have the data. I can recommend https://t.co/HfMGsPtnbD for fine-tuning if you want to give it a go. Or shout if you need a hand.
End result model wil probably only be useful for style transfer, not for content generation.