'Like Spotify for academic articles', the slogan says. It gives access to a claimed 10,000 paywalled academic journals for $40USD/month ($30/month for a year). The site correctly claims that you can therefore get a whole year's access to all these journals for the cost of about 10 individual articles in an average paywalled journal, so it seems like a pretty good deal for any researcher outside academia needing to access more than a handful of papers in closed journals a year.

I had a quick browse, and here are my initial observations:

• There's a fairly decent selection of many of the more significant profiteering journals, albeit some that I read regularly are not there (including, interestingly, some paywalled but not-for-profit publications). It's worth noting that, unlike Spotify for music or Netflix for movies, it can be a serious problem if a required paper is not available to a researcher. Some is better than none, but I don't think 10,000 journals is anything like enough to make this truly compelling or disruptive.
• The access is a bit variable - not all is full-text, and it looks like there are some notable limitations on what you can do with at least some of the papers (limits on pages you can print per month is a warning sign - this is at best a rental model, the equivalent of streaming).
• The site seems a bit flaky - the search doesn't work very well, and sometimes fails altogether, and it seems to lose session state very easily - but it's mostly a modern, easy-to-use system.
• There are some useful browser add-ins etc that make it easy to hook in things like Scholar.

It's not up there with a good university library. Not even close. Athabasca University Library, for instance, gives access to and indexes about 65,000 journals, albeit including a number that are open-access already. But AU library also gives access to a host of physical books and journals, and a very large number of online books, loads of conference proceedings, an excellent group of skilled information professionals to provide help with finding what you need, and plenty more.  Our undergraduate students get all of that for 6 months, as well as any textbooks needed for their courses, and their course materials, for a grand total of $180CAD ($30/month), paid as a standard resources fee. We do run this at a substantial loss (costs to us were, according to the last set of figures I saw, over $250/student, mainly thanks to immoral textbook pricing) but, even so, $40USD a month for a fraction of the services seems extremely steep to me. 

It would be unfair, though, to call this pricing predatory: I expect the company has been fleeced by the publishers just like everyone else. DeepDyve is just filling a market niche left by the truly predatory publishers that steal publicly funded research, then hold it to ransom in closed, legislatively locked containers to sell back to those that produced it (and others), lining their filthy pockets with obscenely huge profits all the way down the line. DeepDyve reduces the costs for some people, and that's OK, but it's hardly a solution to the bigger problem, and may actually bolster a status quo that is fundamentally corrupt down to its core, because it provides an ongoing revenue stream to publishers that might otherwise be bypassed by 'grey' sources (if you want papers from paywalled journals and books, mail the author!) or 'pirate' sites like Sci Hub or Academic Torrents. The correct answer to the problem is for all of us to stop publishing in closed, profiteering, exploitative journals, to stop letting them steal from us in the first place.