When errors occur, I have to spend money to make corrections. Does your consumption method take users’ feelings into consideration? The content itself has already collapsed because it was not generated according to instructions. As a result, we have to pay for your mistakes. Is this correct?
AI models cost money to run. There will never be a stage where they get everything right every time, so aligning costs across the whole supply chain gives you the power to choose each time if you’d rather pay for AI assistance or drop into the code and work manually
If cost is a constraining factor for you then the sm model is about 3x cheaper per token