Evaluating different quant levels for your use case is a problem you can pretty reliably throw at a coding agent and leave overnight though. At least, it should

Evaluating different quant levels for your use case is a problem you can pretty reliably throw at a coding agent and leave overnight though. At least, it should give you a much smaller shortlist.