For the elections I mentioned, it was at least a few days out. Within 24 hours, every one of them was accurate, not just in terms of the binary win/lose aspect, but also if it would be a close race or not, as the spread often includes wins based on electoral college vote totals. Functionally, all it can really do for you is not make you feel like you need to stay up late biting your nails, and give you an idea of what's really going to happen. Maybe enough time to call people on election day and make sure they voted.

This isn't just me speculating, it's well-studied. https://theconversation.com/joe-biden-how-betting-markets-foresaw-the-result-of-the-2020-us-election-150095

They were better at 5 minute/hour before networks called results, for sure. 2016 had HRC/dems for congress 80% odds of winning both day before, and before polls closed. 2020 has Biden winning during that time, but when results were being announced state by state, Trump jumped to a 75% chance of winning. Here, the networks were right to say "but Philadelphia hasn't counted yet" type of hedges to "fake early leads" by Trump who was yelling to stop counting. 2024 was predicted well the whole time.