It's got no data quality checks, so you need to use it with caution, but something like http://arxiv.org/list/hep-ex/recent is a really nice tool for High Energy Physics (you can see quite a few of the first results from the LHC there, for example, without paying any journal subs.)

But it's a much better academic tool than a commercial one, and I would not be surprised that a lot of cranks put stuff up there and claim respectability by association, especially in theory or mathematics.